projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ixgbe: remove redundant DMA alignment code
[deliverable/linux.git] / drivers / net / ixgbe / ixgbe_main.c
1 /*******************************************************************************
2
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2010 Intel Corporation.
5
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
9
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
21
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
34 #include <linux/in.h>
35 #include <linux/ip.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
44 #include <scsi/fc/fc_fcoe.h>
45
46 #include "ixgbe.h"
47 #include "ixgbe_common.h"
48 #include "ixgbe_dcb_82599.h"
49 #include "ixgbe_sriov.h"
50
51 char ixgbe_driver_name[] = "ixgbe";
52 static const char ixgbe_driver_string[] =
53 "Intel(R) 10 Gigabit PCI Express Network Driver";
54
55 #define DRV_VERSION "2.0.84-k2"
56 const char ixgbe_driver_version[] = DRV_VERSION;
57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
58
59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
60 [board_82598] = &ixgbe_82598_info,
61 [board_82599] = &ixgbe_82599_info,
62 };
63
64 /* ixgbe_pci_tbl - PCI Device ID Table
65 *
66 * Wildcard entries (PCI_ANY_ID) should come last
67 * Last entry must be all 0s
68 *
69 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
70 * Class, Class Mask, private data (not used) }
71 */
72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
74 board_82598 },
75 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
76 board_82598 },
77 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
78 board_82598 },
79 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
80 board_82598 },
81 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
82 board_82598 },
83 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
84 board_82598 },
85 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
86 board_82598 },
87 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
88 board_82598 },
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
90 board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
92 board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
94 board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
96 board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
98 board_82599 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
100 board_82599 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
102 board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
104 board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
106 board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
108 board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
110 board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
112 board_82599 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
114 board_82599 },
115
116 /* required last entry */
117 {0, }
118 };
119 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
120
121 #ifdef CONFIG_IXGBE_DCA
122 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
123 void *p);
124 static struct notifier_block dca_notifier = {
125 .notifier_call = ixgbe_notify_dca,
126 .next = NULL,
127 .priority = 0
128 };
129 #endif
130
131 #ifdef CONFIG_PCI_IOV
132 static unsigned int max_vfs;
133 module_param(max_vfs, uint, 0);
134 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
135 "per physical function");
136 #endif /* CONFIG_PCI_IOV */
137
138 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
139 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
140 MODULE_LICENSE("GPL");
141 MODULE_VERSION(DRV_VERSION);
142
143 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
144
145 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
146 {
147 struct ixgbe_hw *hw = &adapter->hw;
148 u32 gcr;
149 u32 gpie;
150 u32 vmdctl;
151
152 #ifdef CONFIG_PCI_IOV
153 /* disable iov and allow time for transactions to clear */
154 pci_disable_sriov(adapter->pdev);
155 #endif
156
157 /* turn off device IOV mode */
158 gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
159 gcr &= ~(IXGBE_GCR_EXT_SRIOV);
160 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
161 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
162 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
163 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
164
165 /* set default pool back to 0 */
166 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
167 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
168 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
169
170 /* take a breather then clean up driver data */
171 msleep(100);
172 if (adapter->vfinfo)
173 kfree(adapter->vfinfo);
174 adapter->vfinfo = NULL;
175
176 adapter->num_vfs = 0;
177 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
178 }
179
180 struct ixgbe_reg_info {
181 u32 ofs;
182 char *name;
183 };
184
185 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
186
187 /* General Registers */
188 {IXGBE_CTRL, "CTRL"},
189 {IXGBE_STATUS, "STATUS"},
190 {IXGBE_CTRL_EXT, "CTRL_EXT"},
191
192 /* Interrupt Registers */
193 {IXGBE_EICR, "EICR"},
194
195 /* RX Registers */
196 {IXGBE_SRRCTL(0), "SRRCTL"},
197 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
198 {IXGBE_RDLEN(0), "RDLEN"},
199 {IXGBE_RDH(0), "RDH"},
200 {IXGBE_RDT(0), "RDT"},
201 {IXGBE_RXDCTL(0), "RXDCTL"},
202 {IXGBE_RDBAL(0), "RDBAL"},
203 {IXGBE_RDBAH(0), "RDBAH"},
204
205 /* TX Registers */
206 {IXGBE_TDBAL(0), "TDBAL"},
207 {IXGBE_TDBAH(0), "TDBAH"},
208 {IXGBE_TDLEN(0), "TDLEN"},
209 {IXGBE_TDH(0), "TDH"},
210 {IXGBE_TDT(0), "TDT"},
211 {IXGBE_TXDCTL(0), "TXDCTL"},
212
213 /* List Terminator */
214 {}
215 };
216
217
218 /*
219 * ixgbe_regdump - register printout routine
220 */
221 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
222 {
223 int i = 0, j = 0;
224 char rname[16];
225 u32 regs[64];
226
227 switch (reginfo->ofs) {
228 case IXGBE_SRRCTL(0):
229 for (i = 0; i < 64; i++)
230 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
231 break;
232 case IXGBE_DCA_RXCTRL(0):
233 for (i = 0; i < 64; i++)
234 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
235 break;
236 case IXGBE_RDLEN(0):
237 for (i = 0; i < 64; i++)
238 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
239 break;
240 case IXGBE_RDH(0):
241 for (i = 0; i < 64; i++)
242 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
243 break;
244 case IXGBE_RDT(0):
245 for (i = 0; i < 64; i++)
246 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
247 break;
248 case IXGBE_RXDCTL(0):
249 for (i = 0; i < 64; i++)
250 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
251 break;
252 case IXGBE_RDBAL(0):
253 for (i = 0; i < 64; i++)
254 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
255 break;
256 case IXGBE_RDBAH(0):
257 for (i = 0; i < 64; i++)
258 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
259 break;
260 case IXGBE_TDBAL(0):
261 for (i = 0; i < 64; i++)
262 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
263 break;
264 case IXGBE_TDBAH(0):
265 for (i = 0; i < 64; i++)
266 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
267 break;
268 case IXGBE_TDLEN(0):
269 for (i = 0; i < 64; i++)
270 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
271 break;
272 case IXGBE_TDH(0):
273 for (i = 0; i < 64; i++)
274 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
275 break;
276 case IXGBE_TDT(0):
277 for (i = 0; i < 64; i++)
278 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
279 break;
280 case IXGBE_TXDCTL(0):
281 for (i = 0; i < 64; i++)
282 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
283 break;
284 default:
285 printk(KERN_INFO "%-15s %08x\n", reginfo->name,
286 IXGBE_READ_REG(hw, reginfo->ofs));
287 return;
288 }
289
290 for (i = 0; i < 8; i++) {
291 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
292 printk(KERN_ERR "%-15s ", rname);
293 for (j = 0; j < 8; j++)
294 printk(KERN_CONT "%08x ", regs[i*8+j]);
295 printk(KERN_CONT "\n");
296 }
297
298 }
299
300 /*
301 * ixgbe_dump - Print registers, tx-rings and rx-rings
302 */
303 static void ixgbe_dump(struct ixgbe_adapter *adapter)
304 {
305 struct net_device *netdev = adapter->netdev;
306 struct ixgbe_hw *hw = &adapter->hw;
307 struct ixgbe_reg_info *reginfo;
308 int n = 0;
309 struct ixgbe_ring *tx_ring;
310 struct ixgbe_tx_buffer *tx_buffer_info;
311 union ixgbe_adv_tx_desc *tx_desc;
312 struct my_u0 { u64 a; u64 b; } *u0;
313 struct ixgbe_ring *rx_ring;
314 union ixgbe_adv_rx_desc *rx_desc;
315 struct ixgbe_rx_buffer *rx_buffer_info;
316 u32 staterr;
317 int i = 0;
318
319 if (!netif_msg_hw(adapter))
320 return;
321
322 /* Print netdevice Info */
323 if (netdev) {
324 dev_info(&adapter->pdev->dev, "Net device Info\n");
325 printk(KERN_INFO "Device Name state "
326 "trans_start last_rx\n");
327 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
328 netdev->name,
329 netdev->state,
330 netdev->trans_start,
331 netdev->last_rx);
332 }
333
334 /* Print Registers */
335 dev_info(&adapter->pdev->dev, "Register Dump\n");
336 printk(KERN_INFO " Register Name Value\n");
337 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
338 reginfo->name; reginfo++) {
339 ixgbe_regdump(hw, reginfo);
340 }
341
342 /* Print TX Ring Summary */
343 if (!netdev || !netif_running(netdev))
344 goto exit;
345
346 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
347 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ] "
348 "leng ntw timestamp\n");
349 for (n = 0; n < adapter->num_tx_queues; n++) {
350 tx_ring = adapter->tx_ring[n];
351 tx_buffer_info =
352 &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
353 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
354 n, tx_ring->next_to_use, tx_ring->next_to_clean,
355 (u64)tx_buffer_info->dma,
356 tx_buffer_info->length,
357 tx_buffer_info->next_to_watch,
358 (u64)tx_buffer_info->time_stamp);
359 }
360
361 /* Print TX Rings */
362 if (!netif_msg_tx_done(adapter))
363 goto rx_ring_summary;
364
365 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
366
367 /* Transmit Descriptor Formats
368 *
369 * Advanced Transmit Descriptor
370 * +--------------------------------------------------------------+
371 * 0 | Buffer Address [63:0] |
372 * +--------------------------------------------------------------+
373 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
374 * +--------------------------------------------------------------+
375 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
376 */
377
378 for (n = 0; n < adapter->num_tx_queues; n++) {
379 tx_ring = adapter->tx_ring[n];
380 printk(KERN_INFO "------------------------------------\n");
381 printk(KERN_INFO "TX QUEUE INDEX = %d\n", tx_ring->queue_index);
382 printk(KERN_INFO "------------------------------------\n");
383 printk(KERN_INFO "T [desc] [address 63:0 ] "
384 "[PlPOIdStDDt Ln] [bi->dma ] "
385 "leng ntw timestamp bi->skb\n");
386
387 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
388 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
389 tx_buffer_info = &tx_ring->tx_buffer_info[i];
390 u0 = (struct my_u0 *)tx_desc;
391 printk(KERN_INFO "T [0x%03X] %016llX %016llX %016llX"
392 " %04X %3X %016llX %p", i,
393 le64_to_cpu(u0->a),
394 le64_to_cpu(u0->b),
395 (u64)tx_buffer_info->dma,
396 tx_buffer_info->length,
397 tx_buffer_info->next_to_watch,
398 (u64)tx_buffer_info->time_stamp,
399 tx_buffer_info->skb);
400 if (i == tx_ring->next_to_use &&
401 i == tx_ring->next_to_clean)
402 printk(KERN_CONT " NTC/U\n");
403 else if (i == tx_ring->next_to_use)
404 printk(KERN_CONT " NTU\n");
405 else if (i == tx_ring->next_to_clean)
406 printk(KERN_CONT " NTC\n");
407 else
408 printk(KERN_CONT "\n");
409
410 if (netif_msg_pktdata(adapter) &&
411 tx_buffer_info->dma != 0)
412 print_hex_dump(KERN_INFO, "",
413 DUMP_PREFIX_ADDRESS, 16, 1,
414 phys_to_virt(tx_buffer_info->dma),
415 tx_buffer_info->length, true);
416 }
417 }
418
419 /* Print RX Rings Summary */
420 rx_ring_summary:
421 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
422 printk(KERN_INFO "Queue [NTU] [NTC]\n");
423 for (n = 0; n < adapter->num_rx_queues; n++) {
424 rx_ring = adapter->rx_ring[n];
425 printk(KERN_INFO "%5d %5X %5X\n", n,
426 rx_ring->next_to_use, rx_ring->next_to_clean);
427 }
428
429 /* Print RX Rings */
430 if (!netif_msg_rx_status(adapter))
431 goto exit;
432
433 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
434
435 /* Advanced Receive Descriptor (Read) Format
436 * 63 1 0
437 * +-----------------------------------------------------+
438 * 0 | Packet Buffer Address [63:1] |A0/NSE|
439 * +----------------------------------------------+------+
440 * 8 | Header Buffer Address [63:1] | DD |
441 * +-----------------------------------------------------+
442 *
443 *
444 * Advanced Receive Descriptor (Write-Back) Format
445 *
446 * 63 48 47 32 31 30 21 20 16 15 4 3 0
447 * +------------------------------------------------------+
448 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
449 * | Checksum Ident | | | | Type | Type |
450 * +------------------------------------------------------+
451 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
452 * +------------------------------------------------------+
453 * 63 48 47 32 31 20 19 0
454 */
455 for (n = 0; n < adapter->num_rx_queues; n++) {
456 rx_ring = adapter->rx_ring[n];
457 printk(KERN_INFO "------------------------------------\n");
458 printk(KERN_INFO "RX QUEUE INDEX = %d\n", rx_ring->queue_index);
459 printk(KERN_INFO "------------------------------------\n");
460 printk(KERN_INFO "R [desc] [ PktBuf A0] "
461 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
462 "<-- Adv Rx Read format\n");
463 printk(KERN_INFO "RWB[desc] [PcsmIpSHl PtRs] "
464 "[vl er S cks ln] ---------------- [bi->skb] "
465 "<-- Adv Rx Write-Back format\n");
466
467 for (i = 0; i < rx_ring->count; i++) {
468 rx_buffer_info = &rx_ring->rx_buffer_info[i];
469 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
470 u0 = (struct my_u0 *)rx_desc;
471 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
472 if (staterr & IXGBE_RXD_STAT_DD) {
473 /* Descriptor Done */
474 printk(KERN_INFO "RWB[0x%03X] %016llX "
475 "%016llX ---------------- %p", i,
476 le64_to_cpu(u0->a),
477 le64_to_cpu(u0->b),
478 rx_buffer_info->skb);
479 } else {
480 printk(KERN_INFO "R [0x%03X] %016llX "
481 "%016llX %016llX %p", i,
482 le64_to_cpu(u0->a),
483 le64_to_cpu(u0->b),
484 (u64)rx_buffer_info->dma,
485 rx_buffer_info->skb);
486
487 if (netif_msg_pktdata(adapter)) {
488 print_hex_dump(KERN_INFO, "",
489 DUMP_PREFIX_ADDRESS, 16, 1,
490 phys_to_virt(rx_buffer_info->dma),
491 rx_ring->rx_buf_len, true);
492
493 if (rx_ring->rx_buf_len
494 < IXGBE_RXBUFFER_2048)
495 print_hex_dump(KERN_INFO, "",
496 DUMP_PREFIX_ADDRESS, 16, 1,
497 phys_to_virt(
498 rx_buffer_info->page_dma +
499 rx_buffer_info->page_offset
500 ),
501 PAGE_SIZE/2, true);
502 }
503 }
504
505 if (i == rx_ring->next_to_use)
506 printk(KERN_CONT " NTU\n");
507 else if (i == rx_ring->next_to_clean)
508 printk(KERN_CONT " NTC\n");
509 else
510 printk(KERN_CONT "\n");
511
512 }
513 }
514
515 exit:
516 return;
517 }
518
519 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
520 {
521 u32 ctrl_ext;
522
523 /* Let firmware take over control of h/w */
524 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
525 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
526 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
527 }
528
529 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
530 {
531 u32 ctrl_ext;
532
533 /* Let firmware know the driver has taken over */
534 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
535 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
536 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
537 }
538
539 /*
540 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
541 * @adapter: pointer to adapter struct
542 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
543 * @queue: queue to map the corresponding interrupt to
544 * @msix_vector: the vector to map to the corresponding queue
545 *
546 */
547 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
548 u8 queue, u8 msix_vector)
549 {
550 u32 ivar, index;
551 struct ixgbe_hw *hw = &adapter->hw;
552 switch (hw->mac.type) {
553 case ixgbe_mac_82598EB:
554 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
555 if (direction == -1)
556 direction = 0;
557 index = (((direction * 64) + queue) >> 2) & 0x1F;
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
559 ivar &= ~(0xFF << (8 * (queue & 0x3)));
560 ivar |= (msix_vector << (8 * (queue & 0x3)));
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
562 break;
563 case ixgbe_mac_82599EB:
564 if (direction == -1) {
565 /* other causes */
566 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
567 index = ((queue & 1) * 8);
568 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
569 ivar &= ~(0xFF << index);
570 ivar |= (msix_vector << index);
571 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
572 break;
573 } else {
574 /* tx or rx causes */
575 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
576 index = ((16 * (queue & 1)) + (8 * direction));
577 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
578 ivar &= ~(0xFF << index);
579 ivar |= (msix_vector << index);
580 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
581 break;
582 }
583 default:
584 break;
585 }
586 }
587
588 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
589 u64 qmask)
590 {
591 u32 mask;
592
593 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
594 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
596 } else {
597 mask = (qmask & 0xFFFFFFFF);
598 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
599 mask = (qmask >> 32);
600 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
601 }
602 }
603
604 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
605 struct ixgbe_tx_buffer
606 *tx_buffer_info)
607 {
608 if (tx_buffer_info->dma) {
609 if (tx_buffer_info->mapped_as_page)
610 dma_unmap_page(&adapter->pdev->dev,
611 tx_buffer_info->dma,
612 tx_buffer_info->length,
613 DMA_TO_DEVICE);
614 else
615 dma_unmap_single(&adapter->pdev->dev,
616 tx_buffer_info->dma,
617 tx_buffer_info->length,
618 DMA_TO_DEVICE);
619 tx_buffer_info->dma = 0;
620 }
621 if (tx_buffer_info->skb) {
622 dev_kfree_skb_any(tx_buffer_info->skb);
623 tx_buffer_info->skb = NULL;
624 }
625 tx_buffer_info->time_stamp = 0;
626 /* tx_buffer_info must be completely set up in the transmit path */
627 }
628
629 /**
630 * ixgbe_tx_xon_state - check the tx ring xon state
631 * @adapter: the ixgbe adapter
632 * @tx_ring: the corresponding tx_ring
633 *
634 * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
635 * corresponding TC of this tx_ring when checking TFCS.
636 *
637 * Returns : true if in xon state (currently not paused)
638 */
639 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
640 struct ixgbe_ring *tx_ring)
641 {
642 u32 txoff = IXGBE_TFCS_TXOFF;
643
644 #ifdef CONFIG_IXGBE_DCB
645 if (adapter->dcb_cfg.pfc_mode_enable) {
646 int tc;
647 int reg_idx = tx_ring->reg_idx;
648 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
649
650 switch (adapter->hw.mac.type) {
651 case ixgbe_mac_82598EB:
652 tc = reg_idx >> 2;
653 txoff = IXGBE_TFCS_TXOFF0;
654 break;
655 case ixgbe_mac_82599EB:
656 tc = 0;
657 txoff = IXGBE_TFCS_TXOFF;
658 if (dcb_i == 8) {
659 /* TC0, TC1 */
660 tc = reg_idx >> 5;
661 if (tc == 2) /* TC2, TC3 */
662 tc += (reg_idx - 64) >> 4;
663 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
664 tc += 1 + ((reg_idx - 96) >> 3);
665 } else if (dcb_i == 4) {
666 /* TC0, TC1 */
667 tc = reg_idx >> 6;
668 if (tc == 1) {
669 tc += (reg_idx - 64) >> 5;
670 if (tc == 2) /* TC2, TC3 */
671 tc += (reg_idx - 96) >> 4;
672 }
673 }
674 break;
675 default:
676 tc = 0;
677 }
678 txoff <<= tc;
679 }
680 #endif
681 return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
682 }
683
684 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
685 struct ixgbe_ring *tx_ring,
686 unsigned int eop)
687 {
688 struct ixgbe_hw *hw = &adapter->hw;
689
690 /* Detect a transmit hang in hardware, this serializes the
691 * check with the clearing of time_stamp and movement of eop */
692 adapter->detect_tx_hung = false;
693 if (tx_ring->tx_buffer_info[eop].time_stamp &&
694 time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
695 ixgbe_tx_xon_state(adapter, tx_ring)) {
696 /* detected Tx unit hang */
697 union ixgbe_adv_tx_desc *tx_desc;
698 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
699 e_err(drv, "Detected Tx Unit Hang\n"
700 " Tx Queue <%d>\n"
701 " TDH, TDT <%x>, <%x>\n"
702 " next_to_use <%x>\n"
703 " next_to_clean <%x>\n"
704 "tx_buffer_info[next_to_clean]\n"
705 " time_stamp <%lx>\n"
706 " jiffies <%lx>\n",
707 tx_ring->queue_index,
708 IXGBE_READ_REG(hw, tx_ring->head),
709 IXGBE_READ_REG(hw, tx_ring->tail),
710 tx_ring->next_to_use, eop,
711 tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
712 return true;
713 }
714
715 return false;
716 }
717
718 #define IXGBE_MAX_TXD_PWR 14
719 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
720
721 /* Tx Descriptors needed, worst case */
722 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
723 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
724 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
725 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
726
727 static void ixgbe_tx_timeout(struct net_device *netdev);
728
729 /**
730 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
731 * @q_vector: structure containing interrupt and ring information
732 * @tx_ring: tx ring to clean
733 **/
734 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
735 struct ixgbe_ring *tx_ring)
736 {
737 struct ixgbe_adapter *adapter = q_vector->adapter;
738 struct net_device *netdev = adapter->netdev;
739 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
740 struct ixgbe_tx_buffer *tx_buffer_info;
741 unsigned int i, eop, count = 0;
742 unsigned int total_bytes = 0, total_packets = 0;
743
744 i = tx_ring->next_to_clean;
745 eop = tx_ring->tx_buffer_info[i].next_to_watch;
746 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
747
748 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
749 (count < tx_ring->work_limit)) {
750 bool cleaned = false;
751 rmb(); /* read buffer_info after eop_desc */
752 for ( ; !cleaned; count++) {
753 struct sk_buff *skb;
754 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
755 tx_buffer_info = &tx_ring->tx_buffer_info[i];
756 cleaned = (i == eop);
757 skb = tx_buffer_info->skb;
758
759 if (cleaned && skb) {
760 unsigned int segs, bytecount;
761 unsigned int hlen = skb_headlen(skb);
762
763 /* gso_segs is currently only valid for tcp */
764 segs = skb_shinfo(skb)->gso_segs ?: 1;
765 #ifdef IXGBE_FCOE
766 /* adjust for FCoE Sequence Offload */
767 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
768 && (skb->protocol == htons(ETH_P_FCOE)) &&
769 skb_is_gso(skb)) {
770 hlen = skb_transport_offset(skb) +
771 sizeof(struct fc_frame_header) +
772 sizeof(struct fcoe_crc_eof);
773 segs = DIV_ROUND_UP(skb->len - hlen,
774 skb_shinfo(skb)->gso_size);
775 }
776 #endif /* IXGBE_FCOE */
777 /* multiply data chunks by size of headers */
778 bytecount = ((segs - 1) * hlen) + skb->len;
779 total_packets += segs;
780 total_bytes += bytecount;
781 }
782
783 ixgbe_unmap_and_free_tx_resource(adapter,
784 tx_buffer_info);
785
786 tx_desc->wb.status = 0;
787
788 i++;
789 if (i == tx_ring->count)
790 i = 0;
791 }
792
793 eop = tx_ring->tx_buffer_info[i].next_to_watch;
794 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
795 }
796
797 tx_ring->next_to_clean = i;
798
799 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
800 if (unlikely(count && netif_carrier_ok(netdev) &&
801 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
802 /* Make sure that anybody stopping the queue after this
803 * sees the new next_to_clean.
804 */
805 smp_mb();
806 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
807 !test_bit(__IXGBE_DOWN, &adapter->state)) {
808 netif_wake_subqueue(netdev, tx_ring->queue_index);
809 ++tx_ring->restart_queue;
810 }
811 }
812
813 if (adapter->detect_tx_hung) {
814 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
815 /* schedule immediate reset if we believe we hung */
816 e_info(probe, "tx hang %d detected, resetting "
817 "adapter\n", adapter->tx_timeout_count + 1);
818 ixgbe_tx_timeout(adapter->netdev);
819 }
820 }
821
822 /* re-arm the interrupt */
823 if (count >= tx_ring->work_limit)
824 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
825
826 tx_ring->total_bytes += total_bytes;
827 tx_ring->total_packets += total_packets;
828 tx_ring->stats.packets += total_packets;
829 tx_ring->stats.bytes += total_bytes;
830 return (count < tx_ring->work_limit);
831 }
832
833 #ifdef CONFIG_IXGBE_DCA
834 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
835 struct ixgbe_ring *rx_ring)
836 {
837 u32 rxctrl;
838 int cpu = get_cpu();
839 int q = rx_ring->reg_idx;
840
841 if (rx_ring->cpu != cpu) {
842 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
843 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
844 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
845 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
846 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
847 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
848 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
849 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
850 }
851 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
852 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
853 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
854 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
855 IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
857 rx_ring->cpu = cpu;
858 }
859 put_cpu();
860 }
861
862 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
863 struct ixgbe_ring *tx_ring)
864 {
865 u32 txctrl;
866 int cpu = get_cpu();
867 int q = tx_ring->reg_idx;
868 struct ixgbe_hw *hw = &adapter->hw;
869
870 if (tx_ring->cpu != cpu) {
871 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
872 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
873 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
874 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
875 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
876 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
877 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
878 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
879 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
880 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
881 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
882 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
883 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
884 }
885 tx_ring->cpu = cpu;
886 }
887 put_cpu();
888 }
889
890 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
891 {
892 int i;
893
894 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
895 return;
896
897 /* always use CB2 mode, difference is masked in the CB driver */
898 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
899
900 for (i = 0; i < adapter->num_tx_queues; i++) {
901 adapter->tx_ring[i]->cpu = -1;
902 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
903 }
904 for (i = 0; i < adapter->num_rx_queues; i++) {
905 adapter->rx_ring[i]->cpu = -1;
906 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
907 }
908 }
909
910 static int __ixgbe_notify_dca(struct device *dev, void *data)
911 {
912 struct net_device *netdev = dev_get_drvdata(dev);
913 struct ixgbe_adapter *adapter = netdev_priv(netdev);
914 unsigned long event = *(unsigned long *)data;
915
916 switch (event) {
917 case DCA_PROVIDER_ADD:
918 /* if we're already enabled, don't do it again */
919 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
920 break;
921 if (dca_add_requester(dev) == 0) {
922 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
923 ixgbe_setup_dca(adapter);
924 break;
925 }
926 /* Fall Through since DCA is disabled. */
927 case DCA_PROVIDER_REMOVE:
928 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
929 dca_remove_requester(dev);
930 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
931 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
932 }
933 break;
934 }
935
936 return 0;
937 }
938
939 #endif /* CONFIG_IXGBE_DCA */
940 /**
941 * ixgbe_receive_skb - Send a completed packet up the stack
942 * @adapter: board private structure
943 * @skb: packet to send up
944 * @status: hardware indication of status of receive
945 * @rx_ring: rx descriptor ring (for a specific queue) to setup
946 * @rx_desc: rx descriptor
947 **/
948 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
949 struct sk_buff *skb, u8 status,
950 struct ixgbe_ring *ring,
951 union ixgbe_adv_rx_desc *rx_desc)
952 {
953 struct ixgbe_adapter *adapter = q_vector->adapter;
954 struct napi_struct *napi = &q_vector->napi;
955 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
956 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
957
958 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
959 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
960 vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
961 else
962 napi_gro_receive(napi, skb);
963 } else {
964 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
965 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
966 else
967 netif_rx(skb);
968 }
969 }
970
971 /**
972 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
973 * @adapter: address of board private structure
974 * @status_err: hardware indication of status of receive
975 * @skb: skb currently being received and modified
976 **/
977 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
978 union ixgbe_adv_rx_desc *rx_desc,
979 struct sk_buff *skb)
980 {
981 u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
982
983 skb->ip_summed = CHECKSUM_NONE;
984
985 /* Rx csum disabled */
986 if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
987 return;
988
989 /* if IP and error */
990 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
991 (status_err & IXGBE_RXDADV_ERR_IPE)) {
992 adapter->hw_csum_rx_error++;
993 return;
994 }
995
996 if (!(status_err & IXGBE_RXD_STAT_L4CS))
997 return;
998
999 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1000 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1001
1002 /*
1003 * 82599 errata, UDP frames with a 0 checksum can be marked as
1004 * checksum errors.
1005 */
1006 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1007 (adapter->hw.mac.type == ixgbe_mac_82599EB))
1008 return;
1009
1010 adapter->hw_csum_rx_error++;
1011 return;
1012 }
1013
1014 /* It must be a TCP or UDP packet with a valid checksum */
1015 skb->ip_summed = CHECKSUM_UNNECESSARY;
1016 }
1017
1018 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
1019 struct ixgbe_ring *rx_ring, u32 val)
1020 {
1021 /*
1022 * Force memory writes to complete before letting h/w
1023 * know there are new descriptors to fetch. (Only
1024 * applicable for weak-ordered memory model archs,
1025 * such as IA-64).
1026 */
1027 wmb();
1028 IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
1029 }
1030
1031 /**
1032 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1033 * @adapter: address of board private structure
1034 **/
1035 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
1036 struct ixgbe_ring *rx_ring,
1037 int cleaned_count)
1038 {
1039 struct net_device *netdev = adapter->netdev;
1040 struct pci_dev *pdev = adapter->pdev;
1041 union ixgbe_adv_rx_desc *rx_desc;
1042 struct ixgbe_rx_buffer *bi;
1043 unsigned int i;
1044 unsigned int bufsz = rx_ring->rx_buf_len;
1045
1046 i = rx_ring->next_to_use;
1047 bi = &rx_ring->rx_buffer_info[i];
1048
1049 while (cleaned_count--) {
1050 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1051
1052 if (!bi->page_dma &&
1053 (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1054 if (!bi->page) {
1055 bi->page = netdev_alloc_page(netdev);
1056 if (!bi->page) {
1057 adapter->alloc_rx_page_failed++;
1058 goto no_buffers;
1059 }
1060 bi->page_offset = 0;
1061 } else {
1062 /* use a half page if we're re-using */
1063 bi->page_offset ^= (PAGE_SIZE / 2);
1064 }
1065
1066 bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1067 bi->page_offset,
1068 (PAGE_SIZE / 2),
1069 DMA_FROM_DEVICE);
1070 }
1071
1072 if (!bi->skb) {
1073 struct sk_buff *skb = netdev_alloc_skb_ip_align(netdev,
1074 bufsz);
1075 bi->skb = skb;
1076
1077 if (!skb) {
1078 adapter->alloc_rx_buff_failed++;
1079 goto no_buffers;
1080 }
1081 /* initialize queue mapping */
1082 skb_record_rx_queue(skb, rx_ring->queue_index);
1083 }
1084
1085 if (!bi->dma) {
1086 bi->dma = dma_map_single(&pdev->dev,
1087 bi->skb->data,
1088 rx_ring->rx_buf_len,
1089 DMA_FROM_DEVICE);
1090 }
1091 /* Refresh the desc even if buffer_addrs didn't change because
1092 * each write-back erases this info. */
1093 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1094 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1095 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1096 } else {
1097 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1098 }
1099
1100 i++;
1101 if (i == rx_ring->count)
1102 i = 0;
1103 bi = &rx_ring->rx_buffer_info[i];
1104 }
1105
1106 no_buffers:
1107 if (rx_ring->next_to_use != i) {
1108 rx_ring->next_to_use = i;
1109 if (i-- == 0)
1110 i = (rx_ring->count - 1);
1111
1112 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1113 }
1114 }
1115
1116 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1117 {
1118 return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1119 }
1120
1121 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1122 {
1123 return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1124 }
1125
1126 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1127 {
1128 return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1129 IXGBE_RXDADV_RSCCNT_MASK) >>
1130 IXGBE_RXDADV_RSCCNT_SHIFT;
1131 }
1132
1133 /**
1134 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1135 * @skb: pointer to the last skb in the rsc queue
1136 * @count: pointer to number of packets coalesced in this context
1137 *
1138 * This function changes a queue full of hw rsc buffers into a completed
1139 * packet. It uses the ->prev pointers to find the first packet and then
1140 * turns it into the frag list owner.
1141 **/
1142 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1143 u64 *count)
1144 {
1145 unsigned int frag_list_size = 0;
1146
1147 while (skb->prev) {
1148 struct sk_buff *prev = skb->prev;
1149 frag_list_size += skb->len;
1150 skb->prev = NULL;
1151 skb = prev;
1152 *count += 1;
1153 }
1154
1155 skb_shinfo(skb)->frag_list = skb->next;
1156 skb->next = NULL;
1157 skb->len += frag_list_size;
1158 skb->data_len += frag_list_size;
1159 skb->truesize += frag_list_size;
1160 return skb;
1161 }
1162
1163 struct ixgbe_rsc_cb {
1164 dma_addr_t dma;
1165 bool delay_unmap;
1166 };
1167
1168 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1169
1170 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1171 struct ixgbe_ring *rx_ring,
1172 int *work_done, int work_to_do)
1173 {
1174 struct ixgbe_adapter *adapter = q_vector->adapter;
1175 struct net_device *netdev = adapter->netdev;
1176 struct pci_dev *pdev = adapter->pdev;
1177 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1178 struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1179 struct sk_buff *skb;
1180 unsigned int i, rsc_count = 0;
1181 u32 len, staterr;
1182 u16 hdr_info;
1183 bool cleaned = false;
1184 int cleaned_count = 0;
1185 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1186 #ifdef IXGBE_FCOE
1187 int ddp_bytes = 0;
1188 #endif /* IXGBE_FCOE */
1189
1190 i = rx_ring->next_to_clean;
1191 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
1192 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1193 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1194
1195 while (staterr & IXGBE_RXD_STAT_DD) {
1196 u32 upper_len = 0;
1197 if (*work_done >= work_to_do)
1198 break;
1199 (*work_done)++;
1200
1201 rmb(); /* read descriptor and rx_buffer_info after status DD */
1202 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1203 hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1204 len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1205 IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1206 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1207 if ((len > IXGBE_RX_HDR_SIZE) ||
1208 (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1209 len = IXGBE_RX_HDR_SIZE;
1210 } else {
1211 len = le16_to_cpu(rx_desc->wb.upper.length);
1212 }
1213
1214 cleaned = true;
1215 skb = rx_buffer_info->skb;
1216 prefetch(skb->data);
1217 rx_buffer_info->skb = NULL;
1218
1219 if (rx_buffer_info->dma) {
1220 if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1221 (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1222 (!(skb->prev))) {
1223 /*
1224 * When HWRSC is enabled, delay unmapping
1225 * of the first packet. It carries the
1226 * header information, HW may still
1227 * access the header after the writeback.
1228 * Only unmap it when EOP is reached
1229 */
1230 IXGBE_RSC_CB(skb)->delay_unmap = true;
1231 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1232 } else {
1233 dma_unmap_single(&pdev->dev,
1234 rx_buffer_info->dma,
1235 rx_ring->rx_buf_len,
1236 DMA_FROM_DEVICE);
1237 }
1238 rx_buffer_info->dma = 0;
1239 skb_put(skb, len);
1240 }
1241
1242 if (upper_len) {
1243 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1244 PAGE_SIZE / 2, DMA_FROM_DEVICE);
1245 rx_buffer_info->page_dma = 0;
1246 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1247 rx_buffer_info->page,
1248 rx_buffer_info->page_offset,
1249 upper_len);
1250
1251 if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1252 (page_count(rx_buffer_info->page) != 1))
1253 rx_buffer_info->page = NULL;
1254 else
1255 get_page(rx_buffer_info->page);
1256
1257 skb->len += upper_len;
1258 skb->data_len += upper_len;
1259 skb->truesize += upper_len;
1260 }
1261
1262 i++;
1263 if (i == rx_ring->count)
1264 i = 0;
1265
1266 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
1267 prefetch(next_rxd);
1268 cleaned_count++;
1269
1270 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1271 rsc_count = ixgbe_get_rsc_count(rx_desc);
1272
1273 if (rsc_count) {
1274 u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1275 IXGBE_RXDADV_NEXTP_SHIFT;
1276 next_buffer = &rx_ring->rx_buffer_info[nextp];
1277 } else {
1278 next_buffer = &rx_ring->rx_buffer_info[i];
1279 }
1280
1281 if (staterr & IXGBE_RXD_STAT_EOP) {
1282 if (skb->prev)
1283 skb = ixgbe_transform_rsc_queue(skb, &(rx_ring->rsc_count));
1284 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1285 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1286 dma_unmap_single(&pdev->dev,
1287 IXGBE_RSC_CB(skb)->dma,
1288 rx_ring->rx_buf_len,
1289 DMA_FROM_DEVICE);
1290 IXGBE_RSC_CB(skb)->dma = 0;
1291 IXGBE_RSC_CB(skb)->delay_unmap = false;
1292 }
1293 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1294 rx_ring->rsc_count += skb_shinfo(skb)->nr_frags;
1295 else
1296 rx_ring->rsc_count++;
1297 rx_ring->rsc_flush++;
1298 }
1299 rx_ring->stats.packets++;
1300 rx_ring->stats.bytes += skb->len;
1301 } else {
1302 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1303 rx_buffer_info->skb = next_buffer->skb;
1304 rx_buffer_info->dma = next_buffer->dma;
1305 next_buffer->skb = skb;
1306 next_buffer->dma = 0;
1307 } else {
1308 skb->next = next_buffer->skb;
1309 skb->next->prev = skb;
1310 }
1311 rx_ring->non_eop_descs++;
1312 goto next_desc;
1313 }
1314
1315 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1316 dev_kfree_skb_irq(skb);
1317 goto next_desc;
1318 }
1319
1320 ixgbe_rx_checksum(adapter, rx_desc, skb);
1321
1322 /* probably a little skewed due to removing CRC */
1323 total_rx_bytes += skb->len;
1324 total_rx_packets++;
1325
1326 skb->protocol = eth_type_trans(skb, adapter->netdev);
1327 #ifdef IXGBE_FCOE
1328 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1329 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1330 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1331 if (!ddp_bytes)
1332 goto next_desc;
1333 }
1334 #endif /* IXGBE_FCOE */
1335 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1336
1337 next_desc:
1338 rx_desc->wb.upper.status_error = 0;
1339
1340 /* return some buffers to hardware, one at a time is too slow */
1341 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1342 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1343 cleaned_count = 0;
1344 }
1345
1346 /* use prefetched values */
1347 rx_desc = next_rxd;
1348 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1349
1350 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1351 }
1352
1353 rx_ring->next_to_clean = i;
1354 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1355
1356 if (cleaned_count)
1357 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1358
1359 #ifdef IXGBE_FCOE
1360 /* include DDPed FCoE data */
1361 if (ddp_bytes > 0) {
1362 unsigned int mss;
1363
1364 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1365 sizeof(struct fc_frame_header) -
1366 sizeof(struct fcoe_crc_eof);
1367 if (mss > 512)
1368 mss &= ~511;
1369 total_rx_bytes += ddp_bytes;
1370 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1371 }
1372 #endif /* IXGBE_FCOE */
1373
1374 rx_ring->total_packets += total_rx_packets;
1375 rx_ring->total_bytes += total_rx_bytes;
1376 netdev->stats.rx_bytes += total_rx_bytes;
1377 netdev->stats.rx_packets += total_rx_packets;
1378
1379 return cleaned;
1380 }
1381
1382 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1383 /**
1384 * ixgbe_configure_msix - Configure MSI-X hardware
1385 * @adapter: board private structure
1386 *
1387 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1388 * interrupts.
1389 **/
1390 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1391 {
1392 struct ixgbe_q_vector *q_vector;
1393 int i, j, q_vectors, v_idx, r_idx;
1394 u32 mask;
1395
1396 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1397
1398 /*
1399 * Populate the IVAR table and set the ITR values to the
1400 * corresponding register.
1401 */
1402 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1403 q_vector = adapter->q_vector[v_idx];
1404 /* XXX for_each_set_bit(...) */
1405 r_idx = find_first_bit(q_vector->rxr_idx,
1406 adapter->num_rx_queues);
1407
1408 for (i = 0; i < q_vector->rxr_count; i++) {
1409 j = adapter->rx_ring[r_idx]->reg_idx;
1410 ixgbe_set_ivar(adapter, 0, j, v_idx);
1411 r_idx = find_next_bit(q_vector->rxr_idx,
1412 adapter->num_rx_queues,
1413 r_idx + 1);
1414 }
1415 r_idx = find_first_bit(q_vector->txr_idx,
1416 adapter->num_tx_queues);
1417
1418 for (i = 0; i < q_vector->txr_count; i++) {
1419 j = adapter->tx_ring[r_idx]->reg_idx;
1420 ixgbe_set_ivar(adapter, 1, j, v_idx);
1421 r_idx = find_next_bit(q_vector->txr_idx,
1422 adapter->num_tx_queues,
1423 r_idx + 1);
1424 }
1425
1426 if (q_vector->txr_count && !q_vector->rxr_count)
1427 /* tx only */
1428 q_vector->eitr = adapter->tx_eitr_param;
1429 else if (q_vector->rxr_count)
1430 /* rx or mixed */
1431 q_vector->eitr = adapter->rx_eitr_param;
1432
1433 ixgbe_write_eitr(q_vector);
1434 }
1435
1436 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1437 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1438 v_idx);
1439 else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1440 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1441 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1442
1443 /* set up to autoclear timer, and the vectors */
1444 mask = IXGBE_EIMS_ENABLE_MASK;
1445 if (adapter->num_vfs)
1446 mask &= ~(IXGBE_EIMS_OTHER |
1447 IXGBE_EIMS_MAILBOX |
1448 IXGBE_EIMS_LSC);
1449 else
1450 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1451 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1452 }
1453
1454 enum latency_range {
1455 lowest_latency = 0,
1456 low_latency = 1,
1457 bulk_latency = 2,
1458 latency_invalid = 255
1459 };
1460
1461 /**
1462 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1463 * @adapter: pointer to adapter
1464 * @eitr: eitr setting (ints per sec) to give last timeslice
1465 * @itr_setting: current throttle rate in ints/second
1466 * @packets: the number of packets during this measurement interval
1467 * @bytes: the number of bytes during this measurement interval
1468 *
1469 * Stores a new ITR value based on packets and byte
1470 * counts during the last interrupt. The advantage of per interrupt
1471 * computation is faster updates and more accurate ITR for the current
1472 * traffic pattern. Constants in this function were computed
1473 * based on theoretical maximum wire speed and thresholds were set based
1474 * on testing data as well as attempting to minimize response time
1475 * while increasing bulk throughput.
1476 * this functionality is controlled by the InterruptThrottleRate module
1477 * parameter (see ixgbe_param.c)
1478 **/
1479 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1480 u32 eitr, u8 itr_setting,
1481 int packets, int bytes)
1482 {
1483 unsigned int retval = itr_setting;
1484 u32 timepassed_us;
1485 u64 bytes_perint;
1486
1487 if (packets == 0)
1488 goto update_itr_done;
1489
1490
1491 /* simple throttlerate management
1492 * 0-20MB/s lowest (100000 ints/s)
1493 * 20-100MB/s low (20000 ints/s)
1494 * 100-1249MB/s bulk (8000 ints/s)
1495 */
1496 /* what was last interrupt timeslice? */
1497 timepassed_us = 1000000/eitr;
1498 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1499
1500 switch (itr_setting) {
1501 case lowest_latency:
1502 if (bytes_perint > adapter->eitr_low)
1503 retval = low_latency;
1504 break;
1505 case low_latency:
1506 if (bytes_perint > adapter->eitr_high)
1507 retval = bulk_latency;
1508 else if (bytes_perint <= adapter->eitr_low)
1509 retval = lowest_latency;
1510 break;
1511 case bulk_latency:
1512 if (bytes_perint <= adapter->eitr_high)
1513 retval = low_latency;
1514 break;
1515 }
1516
1517 update_itr_done:
1518 return retval;
1519 }
1520
1521 /**
1522 * ixgbe_write_eitr - write EITR register in hardware specific way
1523 * @q_vector: structure containing interrupt and ring information
1524 *
1525 * This function is made to be called by ethtool and by the driver
1526 * when it needs to update EITR registers at runtime. Hardware
1527 * specific quirks/differences are taken care of here.
1528 */
1529 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1530 {
1531 struct ixgbe_adapter *adapter = q_vector->adapter;
1532 struct ixgbe_hw *hw = &adapter->hw;
1533 int v_idx = q_vector->v_idx;
1534 u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1535
1536 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1537 /* must write high and low 16 bits to reset counter */
1538 itr_reg |= (itr_reg << 16);
1539 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1540 /*
1541 * 82599 can support a value of zero, so allow it for
1542 * max interrupt rate, but there is an errata where it can
1543 * not be zero with RSC
1544 */
1545 if (itr_reg == 8 &&
1546 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1547 itr_reg = 0;
1548
1549 /*
1550 * set the WDIS bit to not clear the timer bits and cause an
1551 * immediate assertion of the interrupt
1552 */
1553 itr_reg |= IXGBE_EITR_CNT_WDIS;
1554 }
1555 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1556 }
1557
1558 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1559 {
1560 struct ixgbe_adapter *adapter = q_vector->adapter;
1561 u32 new_itr;
1562 u8 current_itr, ret_itr;
1563 int i, r_idx;
1564 struct ixgbe_ring *rx_ring, *tx_ring;
1565
1566 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1567 for (i = 0; i < q_vector->txr_count; i++) {
1568 tx_ring = adapter->tx_ring[r_idx];
1569 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1570 q_vector->tx_itr,
1571 tx_ring->total_packets,
1572 tx_ring->total_bytes);
1573 /* if the result for this queue would decrease interrupt
1574 * rate for this vector then use that result */
1575 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1576 q_vector->tx_itr - 1 : ret_itr);
1577 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1578 r_idx + 1);
1579 }
1580
1581 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1582 for (i = 0; i < q_vector->rxr_count; i++) {
1583 rx_ring = adapter->rx_ring[r_idx];
1584 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1585 q_vector->rx_itr,
1586 rx_ring->total_packets,
1587 rx_ring->total_bytes);
1588 /* if the result for this queue would decrease interrupt
1589 * rate for this vector then use that result */
1590 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1591 q_vector->rx_itr - 1 : ret_itr);
1592 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1593 r_idx + 1);
1594 }
1595
1596 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1597
1598 switch (current_itr) {
1599 /* counts and packets in update_itr are dependent on these numbers */
1600 case lowest_latency:
1601 new_itr = 100000;
1602 break;
1603 case low_latency:
1604 new_itr = 20000; /* aka hwitr = ~200 */
1605 break;
1606 case bulk_latency:
1607 default:
1608 new_itr = 8000;
1609 break;
1610 }
1611
1612 if (new_itr != q_vector->eitr) {
1613 /* do an exponential smoothing */
1614 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1615
1616 /* save the algorithm value here, not the smoothed one */
1617 q_vector->eitr = new_itr;
1618
1619 ixgbe_write_eitr(q_vector);
1620 }
1621 }
1622
1623 /**
1624 * ixgbe_check_overtemp_task - worker thread to check over tempurature
1625 * @work: pointer to work_struct containing our data
1626 **/
1627 static void ixgbe_check_overtemp_task(struct work_struct *work)
1628 {
1629 struct ixgbe_adapter *adapter = container_of(work,
1630 struct ixgbe_adapter,
1631 check_overtemp_task);
1632 struct ixgbe_hw *hw = &adapter->hw;
1633 u32 eicr = adapter->interrupt_event;
1634
1635 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
1636 switch (hw->device_id) {
1637 case IXGBE_DEV_ID_82599_T3_LOM: {
1638 u32 autoneg;
1639 bool link_up = false;
1640
1641 if (hw->mac.ops.check_link)
1642 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1643
1644 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1645 (eicr & IXGBE_EICR_LSC))
1646 /* Check if this is due to overtemp */
1647 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1648 break;
1649 }
1650 return;
1651 default:
1652 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1653 return;
1654 break;
1655 }
1656 e_crit(drv, "Network adapter has been stopped because it has "
1657 "over heated. Restart the computer. If the problem "
1658 "persists, power off the system and replace the "
1659 "adapter\n");
1660 /* write to clear the interrupt */
1661 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1662 }
1663 }
1664
1665 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1666 {
1667 struct ixgbe_hw *hw = &adapter->hw;
1668
1669 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1670 (eicr & IXGBE_EICR_GPI_SDP1)) {
1671 e_crit(probe, "Fan has stopped, replace the adapter\n");
1672 /* write to clear the interrupt */
1673 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1674 }
1675 }
1676
1677 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1678 {
1679 struct ixgbe_hw *hw = &adapter->hw;
1680
1681 if (eicr & IXGBE_EICR_GPI_SDP1) {
1682 /* Clear the interrupt */
1683 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1684 schedule_work(&adapter->multispeed_fiber_task);
1685 } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1686 /* Clear the interrupt */
1687 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1688 schedule_work(&adapter->sfp_config_module_task);
1689 } else {
1690 /* Interrupt isn't for us... */
1691 return;
1692 }
1693 }
1694
1695 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1696 {
1697 struct ixgbe_hw *hw = &adapter->hw;
1698
1699 adapter->lsc_int++;
1700 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1701 adapter->link_check_timeout = jiffies;
1702 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1703 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1704 IXGBE_WRITE_FLUSH(hw);
1705 schedule_work(&adapter->watchdog_task);
1706 }
1707 }
1708
1709 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1710 {
1711 struct net_device *netdev = data;
1712 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1713 struct ixgbe_hw *hw = &adapter->hw;
1714 u32 eicr;
1715
1716 /*
1717 * Workaround for Silicon errata. Use clear-by-write instead
1718 * of clear-by-read. Reading with EICS will return the
1719 * interrupt causes without clearing, which later be done
1720 * with the write to EICR.
1721 */
1722 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1723 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1724
1725 if (eicr & IXGBE_EICR_LSC)
1726 ixgbe_check_lsc(adapter);
1727
1728 if (eicr & IXGBE_EICR_MAILBOX)
1729 ixgbe_msg_task(adapter);
1730
1731 if (hw->mac.type == ixgbe_mac_82598EB)
1732 ixgbe_check_fan_failure(adapter, eicr);
1733
1734 if (hw->mac.type == ixgbe_mac_82599EB) {
1735 ixgbe_check_sfp_event(adapter, eicr);
1736 adapter->interrupt_event = eicr;
1737 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1738 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1739 schedule_work(&adapter->check_overtemp_task);
1740
1741 /* Handle Flow Director Full threshold interrupt */
1742 if (eicr & IXGBE_EICR_FLOW_DIR) {
1743 int i;
1744 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1745 /* Disable transmits before FDIR Re-initialization */
1746 netif_tx_stop_all_queues(netdev);
1747 for (i = 0; i < adapter->num_tx_queues; i++) {
1748 struct ixgbe_ring *tx_ring =
1749 adapter->tx_ring[i];
1750 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1751 &tx_ring->reinit_state))
1752 schedule_work(&adapter->fdir_reinit_task);
1753 }
1754 }
1755 }
1756 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1757 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1758
1759 return IRQ_HANDLED;
1760 }
1761
1762 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1763 u64 qmask)
1764 {
1765 u32 mask;
1766
1767 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1768 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1769 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1770 } else {
1771 mask = (qmask & 0xFFFFFFFF);
1772 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1773 mask = (qmask >> 32);
1774 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1775 }
1776 /* skip the flush */
1777 }
1778
1779 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1780 u64 qmask)
1781 {
1782 u32 mask;
1783
1784 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1785 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1786 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1787 } else {
1788 mask = (qmask & 0xFFFFFFFF);
1789 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1790 mask = (qmask >> 32);
1791 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1792 }
1793 /* skip the flush */
1794 }
1795
1796 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1797 {
1798 struct ixgbe_q_vector *q_vector = data;
1799 struct ixgbe_adapter *adapter = q_vector->adapter;
1800 struct ixgbe_ring *tx_ring;
1801 int i, r_idx;
1802
1803 if (!q_vector->txr_count)
1804 return IRQ_HANDLED;
1805
1806 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1807 for (i = 0; i < q_vector->txr_count; i++) {
1808 tx_ring = adapter->tx_ring[r_idx];
1809 tx_ring->total_bytes = 0;
1810 tx_ring->total_packets = 0;
1811 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1812 r_idx + 1);
1813 }
1814
1815 /* EIAM disabled interrupts (on this vector) for us */
1816 napi_schedule(&q_vector->napi);
1817
1818 return IRQ_HANDLED;
1819 }
1820
1821 /**
1822 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1823 * @irq: unused
1824 * @data: pointer to our q_vector struct for this interrupt vector
1825 **/
1826 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1827 {
1828 struct ixgbe_q_vector *q_vector = data;
1829 struct ixgbe_adapter *adapter = q_vector->adapter;
1830 struct ixgbe_ring *rx_ring;
1831 int r_idx;
1832 int i;
1833
1834 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1835 for (i = 0; i < q_vector->rxr_count; i++) {
1836 rx_ring = adapter->rx_ring[r_idx];
1837 rx_ring->total_bytes = 0;
1838 rx_ring->total_packets = 0;
1839 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1840 r_idx + 1);
1841 }
1842
1843 if (!q_vector->rxr_count)
1844 return IRQ_HANDLED;
1845
1846 /* disable interrupts on this vector only */
1847 /* EIAM disabled interrupts (on this vector) for us */
1848 napi_schedule(&q_vector->napi);
1849
1850 return IRQ_HANDLED;
1851 }
1852
1853 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1854 {
1855 struct ixgbe_q_vector *q_vector = data;
1856 struct ixgbe_adapter *adapter = q_vector->adapter;
1857 struct ixgbe_ring *ring;
1858 int r_idx;
1859 int i;
1860
1861 if (!q_vector->txr_count && !q_vector->rxr_count)
1862 return IRQ_HANDLED;
1863
1864 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1865 for (i = 0; i < q_vector->txr_count; i++) {
1866 ring = adapter->tx_ring[r_idx];
1867 ring->total_bytes = 0;
1868 ring->total_packets = 0;
1869 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1870 r_idx + 1);
1871 }
1872
1873 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1874 for (i = 0; i < q_vector->rxr_count; i++) {
1875 ring = adapter->rx_ring[r_idx];
1876 ring->total_bytes = 0;
1877 ring->total_packets = 0;
1878 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1879 r_idx + 1);
1880 }
1881
1882 /* EIAM disabled interrupts (on this vector) for us */
1883 napi_schedule(&q_vector->napi);
1884
1885 return IRQ_HANDLED;
1886 }
1887
1888 /**
1889 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1890 * @napi: napi struct with our devices info in it
1891 * @budget: amount of work driver is allowed to do this pass, in packets
1892 *
1893 * This function is optimized for cleaning one queue only on a single
1894 * q_vector!!!
1895 **/
1896 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1897 {
1898 struct ixgbe_q_vector *q_vector =
1899 container_of(napi, struct ixgbe_q_vector, napi);
1900 struct ixgbe_adapter *adapter = q_vector->adapter;
1901 struct ixgbe_ring *rx_ring = NULL;
1902 int work_done = 0;
1903 long r_idx;
1904
1905 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1906 rx_ring = adapter->rx_ring[r_idx];
1907 #ifdef CONFIG_IXGBE_DCA
1908 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1909 ixgbe_update_rx_dca(adapter, rx_ring);
1910 #endif
1911
1912 ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1913
1914 /* If all Rx work done, exit the polling mode */
1915 if (work_done < budget) {
1916 napi_complete(napi);
1917 if (adapter->rx_itr_setting & 1)
1918 ixgbe_set_itr_msix(q_vector);
1919 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1920 ixgbe_irq_enable_queues(adapter,
1921 ((u64)1 << q_vector->v_idx));
1922 }
1923
1924 return work_done;
1925 }
1926
1927 /**
1928 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1929 * @napi: napi struct with our devices info in it
1930 * @budget: amount of work driver is allowed to do this pass, in packets
1931 *
1932 * This function will clean more than one rx queue associated with a
1933 * q_vector.
1934 **/
1935 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1936 {
1937 struct ixgbe_q_vector *q_vector =
1938 container_of(napi, struct ixgbe_q_vector, napi);
1939 struct ixgbe_adapter *adapter = q_vector->adapter;
1940 struct ixgbe_ring *ring = NULL;
1941 int work_done = 0, i;
1942 long r_idx;
1943 bool tx_clean_complete = true;
1944
1945 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1946 for (i = 0; i < q_vector->txr_count; i++) {
1947 ring = adapter->tx_ring[r_idx];
1948 #ifdef CONFIG_IXGBE_DCA
1949 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1950 ixgbe_update_tx_dca(adapter, ring);
1951 #endif
1952 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1953 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1954 r_idx + 1);
1955 }
1956
1957 /* attempt to distribute budget to each queue fairly, but don't allow
1958 * the budget to go below 1 because we'll exit polling */
1959 budget /= (q_vector->rxr_count ?: 1);
1960 budget = max(budget, 1);
1961 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1962 for (i = 0; i < q_vector->rxr_count; i++) {
1963 ring = adapter->rx_ring[r_idx];
1964 #ifdef CONFIG_IXGBE_DCA
1965 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1966 ixgbe_update_rx_dca(adapter, ring);
1967 #endif
1968 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1969 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1970 r_idx + 1);
1971 }
1972
1973 r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1974 ring = adapter->rx_ring[r_idx];
1975 /* If all Rx work done, exit the polling mode */
1976 if (work_done < budget) {
1977 napi_complete(napi);
1978 if (adapter->rx_itr_setting & 1)
1979 ixgbe_set_itr_msix(q_vector);
1980 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1981 ixgbe_irq_enable_queues(adapter,
1982 ((u64)1 << q_vector->v_idx));
1983 return 0;
1984 }
1985
1986 return work_done;
1987 }
1988
1989 /**
1990 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1991 * @napi: napi struct with our devices info in it
1992 * @budget: amount of work driver is allowed to do this pass, in packets
1993 *
1994 * This function is optimized for cleaning one queue only on a single
1995 * q_vector!!!
1996 **/
1997 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1998 {
1999 struct ixgbe_q_vector *q_vector =
2000 container_of(napi, struct ixgbe_q_vector, napi);
2001 struct ixgbe_adapter *adapter = q_vector->adapter;
2002 struct ixgbe_ring *tx_ring = NULL;
2003 int work_done = 0;
2004 long r_idx;
2005
2006 r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2007 tx_ring = adapter->tx_ring[r_idx];
2008 #ifdef CONFIG_IXGBE_DCA
2009 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2010 ixgbe_update_tx_dca(adapter, tx_ring);
2011 #endif
2012
2013 if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2014 work_done = budget;
2015
2016 /* If all Tx work done, exit the polling mode */
2017 if (work_done < budget) {
2018 napi_complete(napi);
2019 if (adapter->tx_itr_setting & 1)
2020 ixgbe_set_itr_msix(q_vector);
2021 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2022 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2023 }
2024
2025 return work_done;
2026 }
2027
2028 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2029 int r_idx)
2030 {
2031 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2032
2033 set_bit(r_idx, q_vector->rxr_idx);
2034 q_vector->rxr_count++;
2035 }
2036
2037 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2038 int t_idx)
2039 {
2040 struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2041
2042 set_bit(t_idx, q_vector->txr_idx);
2043 q_vector->txr_count++;
2044 }
2045
2046 /**
2047 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2048 * @adapter: board private structure to initialize
2049 * @vectors: allotted vector count for descriptor rings
2050 *
2051 * This function maps descriptor rings to the queue-specific vectors
2052 * we were allotted through the MSI-X enabling code. Ideally, we'd have
2053 * one vector per ring/queue, but on a constrained vector budget, we
2054 * group the rings as "efficiently" as possible. You would add new
2055 * mapping configurations in here.
2056 **/
2057 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2058 int vectors)
2059 {
2060 int v_start = 0;
2061 int rxr_idx = 0, txr_idx = 0;
2062 int rxr_remaining = adapter->num_rx_queues;
2063 int txr_remaining = adapter->num_tx_queues;
2064 int i, j;
2065 int rqpv, tqpv;
2066 int err = 0;
2067
2068 /* No mapping required if MSI-X is disabled. */
2069 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2070 goto out;
2071
2072 /*
2073 * The ideal configuration...
2074 * We have enough vectors to map one per queue.
2075 */
2076 if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2077 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2078 map_vector_to_rxq(adapter, v_start, rxr_idx);
2079
2080 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2081 map_vector_to_txq(adapter, v_start, txr_idx);
2082
2083 goto out;
2084 }
2085
2086 /*
2087 * If we don't have enough vectors for a 1-to-1
2088 * mapping, we'll have to group them so there are
2089 * multiple queues per vector.
2090 */
2091 /* Re-adjusting *qpv takes care of the remainder. */
2092 for (i = v_start; i < vectors; i++) {
2093 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2094 for (j = 0; j < rqpv; j++) {
2095 map_vector_to_rxq(adapter, i, rxr_idx);
2096 rxr_idx++;
2097 rxr_remaining--;
2098 }
2099 }
2100 for (i = v_start; i < vectors; i++) {
2101 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2102 for (j = 0; j < tqpv; j++) {
2103 map_vector_to_txq(adapter, i, txr_idx);
2104 txr_idx++;
2105 txr_remaining--;
2106 }
2107 }
2108
2109 out:
2110 return err;
2111 }
2112
2113 /**
2114 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2115 * @adapter: board private structure
2116 *
2117 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2118 * interrupts from the kernel.
2119 **/
2120 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2121 {
2122 struct net_device *netdev = adapter->netdev;
2123 irqreturn_t (*handler)(int, void *);
2124 int i, vector, q_vectors, err;
2125 int ri=0, ti=0;
2126
2127 /* Decrement for Other and TCP Timer vectors */
2128 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2129
2130 /* Map the Tx/Rx rings to the vectors we were allotted. */
2131 err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2132 if (err)
2133 goto out;
2134
2135 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2136 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2137 &ixgbe_msix_clean_many)
2138 for (vector = 0; vector < q_vectors; vector++) {
2139 handler = SET_HANDLER(adapter->q_vector[vector]);
2140
2141 if(handler == &ixgbe_msix_clean_rx) {
2142 sprintf(adapter->name[vector], "%s-%s-%d",
2143 netdev->name, "rx", ri++);
2144 }
2145 else if(handler == &ixgbe_msix_clean_tx) {
2146 sprintf(adapter->name[vector], "%s-%s-%d",
2147 netdev->name, "tx", ti++);
2148 }
2149 else
2150 sprintf(adapter->name[vector], "%s-%s-%d",
2151 netdev->name, "TxRx", vector);
2152
2153 err = request_irq(adapter->msix_entries[vector].vector,
2154 handler, 0, adapter->name[vector],
2155 adapter->q_vector[vector]);
2156 if (err) {
2157 e_err(probe, "request_irq failed for MSIX interrupt "
2158 "Error: %d\n", err);
2159 goto free_queue_irqs;
2160 }
2161 }
2162
2163 sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2164 err = request_irq(adapter->msix_entries[vector].vector,
2165 ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2166 if (err) {
2167 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2168 goto free_queue_irqs;
2169 }
2170
2171 return 0;
2172
2173 free_queue_irqs:
2174 for (i = vector - 1; i >= 0; i--)
2175 free_irq(adapter->msix_entries[--vector].vector,
2176 adapter->q_vector[i]);
2177 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2178 pci_disable_msix(adapter->pdev);
2179 kfree(adapter->msix_entries);
2180 adapter->msix_entries = NULL;
2181 out:
2182 return err;
2183 }
2184
2185 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2186 {
2187 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2188 u8 current_itr;
2189 u32 new_itr = q_vector->eitr;
2190 struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2191 struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2192
2193 q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2194 q_vector->tx_itr,
2195 tx_ring->total_packets,
2196 tx_ring->total_bytes);
2197 q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2198 q_vector->rx_itr,
2199 rx_ring->total_packets,
2200 rx_ring->total_bytes);
2201
2202 current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2203
2204 switch (current_itr) {
2205 /* counts and packets in update_itr are dependent on these numbers */
2206 case lowest_latency:
2207 new_itr = 100000;
2208 break;
2209 case low_latency:
2210 new_itr = 20000; /* aka hwitr = ~200 */
2211 break;
2212 case bulk_latency:
2213 new_itr = 8000;
2214 break;
2215 default:
2216 break;
2217 }
2218
2219 if (new_itr != q_vector->eitr) {
2220 /* do an exponential smoothing */
2221 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2222
2223 /* save the algorithm value here, not the smoothed one */
2224 q_vector->eitr = new_itr;
2225
2226 ixgbe_write_eitr(q_vector);
2227 }
2228 }
2229
2230 /**
2231 * ixgbe_irq_enable - Enable default interrupt generation settings
2232 * @adapter: board private structure
2233 **/
2234 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
2235 {
2236 u32 mask;
2237
2238 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2239 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2240 mask |= IXGBE_EIMS_GPI_SDP0;
2241 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2242 mask |= IXGBE_EIMS_GPI_SDP1;
2243 if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2244 mask |= IXGBE_EIMS_ECC;
2245 mask |= IXGBE_EIMS_GPI_SDP1;
2246 mask |= IXGBE_EIMS_GPI_SDP2;
2247 if (adapter->num_vfs)
2248 mask |= IXGBE_EIMS_MAILBOX;
2249 }
2250 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2251 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2252 mask |= IXGBE_EIMS_FLOW_DIR;
2253
2254 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2255 ixgbe_irq_enable_queues(adapter, ~0);
2256 IXGBE_WRITE_FLUSH(&adapter->hw);
2257
2258 if (adapter->num_vfs > 32) {
2259 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2260 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2261 }
2262 }
2263
2264 /**
2265 * ixgbe_intr - legacy mode Interrupt Handler
2266 * @irq: interrupt number
2267 * @data: pointer to a network interface device structure
2268 **/
2269 static irqreturn_t ixgbe_intr(int irq, void *data)
2270 {
2271 struct net_device *netdev = data;
2272 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2273 struct ixgbe_hw *hw = &adapter->hw;
2274 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2275 u32 eicr;
2276
2277 /*
2278 * Workaround for silicon errata. Mask the interrupts
2279 * before the read of EICR.
2280 */
2281 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2282
2283 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2284 * therefore no explict interrupt disable is necessary */
2285 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2286 if (!eicr) {
2287 /* shared interrupt alert!
2288 * make sure interrupts are enabled because the read will
2289 * have disabled interrupts due to EIAM */
2290 ixgbe_irq_enable(adapter);
2291 return IRQ_NONE; /* Not our interrupt */
2292 }
2293
2294 if (eicr & IXGBE_EICR_LSC)
2295 ixgbe_check_lsc(adapter);
2296
2297 if (hw->mac.type == ixgbe_mac_82599EB)
2298 ixgbe_check_sfp_event(adapter, eicr);
2299
2300 ixgbe_check_fan_failure(adapter, eicr);
2301 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2302 ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2303 schedule_work(&adapter->check_overtemp_task);
2304
2305 if (napi_schedule_prep(&(q_vector->napi))) {
2306 adapter->tx_ring[0]->total_packets = 0;
2307 adapter->tx_ring[0]->total_bytes = 0;
2308 adapter->rx_ring[0]->total_packets = 0;
2309 adapter->rx_ring[0]->total_bytes = 0;
2310 /* would disable interrupts here but EIAM disabled it */
2311 __napi_schedule(&(q_vector->napi));
2312 }
2313
2314 return IRQ_HANDLED;
2315 }
2316
2317 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2318 {
2319 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2320
2321 for (i = 0; i < q_vectors; i++) {
2322 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2323 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2324 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2325 q_vector->rxr_count = 0;
2326 q_vector->txr_count = 0;
2327 }
2328 }
2329
2330 /**
2331 * ixgbe_request_irq - initialize interrupts
2332 * @adapter: board private structure
2333 *
2334 * Attempts to configure interrupts using the best available
2335 * capabilities of the hardware and kernel.
2336 **/
2337 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2338 {
2339 struct net_device *netdev = adapter->netdev;
2340 int err;
2341
2342 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2343 err = ixgbe_request_msix_irqs(adapter);
2344 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2345 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2346 netdev->name, netdev);
2347 } else {
2348 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2349 netdev->name, netdev);
2350 }
2351
2352 if (err)
2353 e_err(probe, "request_irq failed, Error %d\n", err);
2354
2355 return err;
2356 }
2357
2358 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2359 {
2360 struct net_device *netdev = adapter->netdev;
2361
2362 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2363 int i, q_vectors;
2364
2365 q_vectors = adapter->num_msix_vectors;
2366
2367 i = q_vectors - 1;
2368 free_irq(adapter->msix_entries[i].vector, netdev);
2369
2370 i--;
2371 for (; i >= 0; i--) {
2372 free_irq(adapter->msix_entries[i].vector,
2373 adapter->q_vector[i]);
2374 }
2375
2376 ixgbe_reset_q_vectors(adapter);
2377 } else {
2378 free_irq(adapter->pdev->irq, netdev);
2379 }
2380 }
2381
2382 /**
2383 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2384 * @adapter: board private structure
2385 **/
2386 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2387 {
2388 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2389 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2390 } else {
2391 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2392 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2393 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2394 if (adapter->num_vfs > 32)
2395 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2396 }
2397 IXGBE_WRITE_FLUSH(&adapter->hw);
2398 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2399 int i;
2400 for (i = 0; i < adapter->num_msix_vectors; i++)
2401 synchronize_irq(adapter->msix_entries[i].vector);
2402 } else {
2403 synchronize_irq(adapter->pdev->irq);
2404 }
2405 }
2406
2407 /**
2408 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2409 *
2410 **/
2411 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2412 {
2413 struct ixgbe_hw *hw = &adapter->hw;
2414
2415 IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2416 EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2417
2418 ixgbe_set_ivar(adapter, 0, 0, 0);
2419 ixgbe_set_ivar(adapter, 1, 0, 0);
2420
2421 map_vector_to_rxq(adapter, 0, 0);
2422 map_vector_to_txq(adapter, 0, 0);
2423
2424 e_info(hw, "Legacy interrupt IVAR setup done\n");
2425 }
2426
2427 /**
2428 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2429 * @adapter: board private structure
2430 *
2431 * Configure the Tx unit of the MAC after a reset.
2432 **/
2433 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2434 {
2435 u64 tdba;
2436 struct ixgbe_hw *hw = &adapter->hw;
2437 u32 i, j, tdlen, txctrl;
2438
2439 /* Setup the HW Tx Head and Tail descriptor pointers */
2440 for (i = 0; i < adapter->num_tx_queues; i++) {
2441 struct ixgbe_ring *ring = adapter->tx_ring[i];
2442 j = ring->reg_idx;
2443 tdba = ring->dma;
2444 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
2445 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
2446 (tdba & DMA_BIT_MASK(32)));
2447 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
2448 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
2449 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
2450 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
2451 adapter->tx_ring[i]->head = IXGBE_TDH(j);
2452 adapter->tx_ring[i]->tail = IXGBE_TDT(j);
2453 /*
2454 * Disable Tx Head Writeback RO bit, since this hoses
2455 * bookkeeping if things aren't delivered in order.
2456 */
2457 switch (hw->mac.type) {
2458 case ixgbe_mac_82598EB:
2459 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
2460 break;
2461 case ixgbe_mac_82599EB:
2462 default:
2463 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
2464 break;
2465 }
2466 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
2467 switch (hw->mac.type) {
2468 case ixgbe_mac_82598EB:
2469 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
2470 break;
2471 case ixgbe_mac_82599EB:
2472 default:
2473 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
2474 break;
2475 }
2476 }
2477
2478 if (hw->mac.type == ixgbe_mac_82599EB) {
2479 u32 rttdcs;
2480 u32 mask;
2481
2482 /* disable the arbiter while setting MTQC */
2483 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2484 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2485 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2486
2487 /* set transmit pool layout */
2488 mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2489 switch (adapter->flags & mask) {
2490
2491 case (IXGBE_FLAG_SRIOV_ENABLED):
2492 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2493 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2494 break;
2495
2496 case (IXGBE_FLAG_DCB_ENABLED):
2497 /* We enable 8 traffic classes, DCB only */
2498 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2499 (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2500 break;
2501
2502 default:
2503 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2504 break;
2505 }
2506
2507 /* re-eable the arbiter */
2508 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2509 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2510 }
2511 }
2512
2513 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2514
2515 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2516 struct ixgbe_ring *rx_ring)
2517 {
2518 u32 srrctl;
2519 int index;
2520 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2521
2522 index = rx_ring->reg_idx;
2523 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2524 unsigned long mask;
2525 mask = (unsigned long) feature[RING_F_RSS].mask;
2526 index = index & mask;
2527 }
2528 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2529
2530 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2531 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2532
2533 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2534 IXGBE_SRRCTL_BSIZEHDR_MASK;
2535
2536 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2537 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2538 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2539 #else
2540 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2541 #endif
2542 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2543 } else {
2544 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2545 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2546 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2547 }
2548
2549 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2550 }
2551
2552 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2553 {
2554 u32 mrqc = 0;
2555 int mask;
2556
2557 if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2558 return mrqc;
2559
2560 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2561 #ifdef CONFIG_IXGBE_DCB
2562 | IXGBE_FLAG_DCB_ENABLED
2563 #endif
2564 | IXGBE_FLAG_SRIOV_ENABLED
2565 );
2566
2567 switch (mask) {
2568 case (IXGBE_FLAG_RSS_ENABLED):
2569 mrqc = IXGBE_MRQC_RSSEN;
2570 break;
2571 case (IXGBE_FLAG_SRIOV_ENABLED):
2572 mrqc = IXGBE_MRQC_VMDQEN;
2573 break;
2574 #ifdef CONFIG_IXGBE_DCB
2575 case (IXGBE_FLAG_DCB_ENABLED):
2576 mrqc = IXGBE_MRQC_RT8TCEN;
2577 break;
2578 #endif /* CONFIG_IXGBE_DCB */
2579 default:
2580 break;
2581 }
2582
2583 return mrqc;
2584 }
2585
2586 /**
2587 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2588 * @adapter: address of board private structure
2589 * @index: index of ring to set
2590 **/
2591 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index)
2592 {
2593 struct ixgbe_ring *rx_ring;
2594 struct ixgbe_hw *hw = &adapter->hw;
2595 int j;
2596 u32 rscctrl;
2597 int rx_buf_len;
2598
2599 rx_ring = adapter->rx_ring[index];
2600 j = rx_ring->reg_idx;
2601 rx_buf_len = rx_ring->rx_buf_len;
2602 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2603 rscctrl |= IXGBE_RSCCTL_RSCEN;
2604 /*
2605 * we must limit the number of descriptors so that the
2606 * total size of max desc * buf_len is not greater
2607 * than 65535
2608 */
2609 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2610 #if (MAX_SKB_FRAGS > 16)
2611 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2612 #elif (MAX_SKB_FRAGS > 8)
2613 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2614 #elif (MAX_SKB_FRAGS > 4)
2615 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2616 #else
2617 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2618 #endif
2619 } else {
2620 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2621 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2622 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2623 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2624 else
2625 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2626 }
2627 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2628 }
2629
2630 /**
2631 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2632 * @adapter: board private structure
2633 *
2634 * Configure the Rx unit of the MAC after a reset.
2635 **/
2636 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2637 {
2638 u64 rdba;
2639 struct ixgbe_hw *hw = &adapter->hw;
2640 struct ixgbe_ring *rx_ring;
2641 struct net_device *netdev = adapter->netdev;
2642 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2643 int i, j;
2644 u32 rdlen, rxctrl, rxcsum;
2645 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2646 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2647 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2648 u32 fctrl, hlreg0;
2649 u32 reta = 0, mrqc = 0;
2650 u32 rdrxctl;
2651 int rx_buf_len;
2652
2653 /* Decide whether to use packet split mode or not */
2654 /* Do not use packet split if we're in SR-IOV Mode */
2655 if (!adapter->num_vfs)
2656 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2657
2658 /* Set the RX buffer length according to the mode */
2659 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2660 rx_buf_len = IXGBE_RX_HDR_SIZE;
2661 if (hw->mac.type == ixgbe_mac_82599EB) {
2662 /* PSRTYPE must be initialized in 82599 */
2663 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2664 IXGBE_PSRTYPE_UDPHDR |
2665 IXGBE_PSRTYPE_IPV4HDR |
2666 IXGBE_PSRTYPE_IPV6HDR |
2667 IXGBE_PSRTYPE_L2HDR;
2668 IXGBE_WRITE_REG(hw,
2669 IXGBE_PSRTYPE(adapter->num_vfs),
2670 psrtype);
2671 }
2672 } else {
2673 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2674 (netdev->mtu <= ETH_DATA_LEN))
2675 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2676 else
2677 rx_buf_len = ALIGN(max_frame, 1024);
2678 }
2679
2680 fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2681 fctrl |= IXGBE_FCTRL_BAM;
2682 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2683 fctrl |= IXGBE_FCTRL_PMCF;
2684 IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2685
2686 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2687 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2688 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2689 else
2690 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2691 #ifdef IXGBE_FCOE
2692 if (netdev->features & NETIF_F_FCOE_MTU)
2693 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2694 #endif
2695 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2696
2697 rdlen = adapter->rx_ring[0]->count * sizeof(union ixgbe_adv_rx_desc);
2698 /* disable receives while setting up the descriptors */
2699 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2700 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2701
2702 /*
2703 * Setup the HW Rx Head and Tail Descriptor Pointers and
2704 * the Base and Length of the Rx Descriptor Ring
2705 */
2706 for (i = 0; i < adapter->num_rx_queues; i++) {
2707 rx_ring = adapter->rx_ring[i];
2708 rdba = rx_ring->dma;
2709 j = rx_ring->reg_idx;
2710 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2711 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2712 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2713 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2714 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2715 rx_ring->head = IXGBE_RDH(j);
2716 rx_ring->tail = IXGBE_RDT(j);
2717 rx_ring->rx_buf_len = rx_buf_len;
2718
2719 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2720 rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2721 else
2722 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2723
2724 #ifdef IXGBE_FCOE
2725 if (netdev->features & NETIF_F_FCOE_MTU) {
2726 struct ixgbe_ring_feature *f;
2727 f = &adapter->ring_feature[RING_F_FCOE];
2728 if ((i >= f->mask) && (i < f->mask + f->indices)) {
2729 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2730 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2731 rx_ring->rx_buf_len =
2732 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2733 }
2734 }
2735
2736 #endif /* IXGBE_FCOE */
2737 ixgbe_configure_srrctl(adapter, rx_ring);
2738 }
2739
2740 if (hw->mac.type == ixgbe_mac_82598EB) {
2741 /*
2742 * For VMDq support of different descriptor types or
2743 * buffer sizes through the use of multiple SRRCTL
2744 * registers, RDRXCTL.MVMEN must be set to 1
2745 *
2746 * also, the manual doesn't mention it clearly but DCA hints
2747 * will only use queue 0's tags unless this bit is set. Side
2748 * effects of setting this bit are only that SRRCTL must be
2749 * fully programmed [0..15]
2750 */
2751 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2752 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2753 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2754 }
2755
2756 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2757 u32 vt_reg_bits;
2758 u32 reg_offset, vf_shift;
2759 u32 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2760 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN
2761 | IXGBE_VT_CTL_REPLEN;
2762 vt_reg_bits |= (adapter->num_vfs <<
2763 IXGBE_VT_CTL_POOL_SHIFT);
2764 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2765 IXGBE_WRITE_REG(hw, IXGBE_MRQC, 0);
2766
2767 vf_shift = adapter->num_vfs % 32;
2768 reg_offset = adapter->num_vfs / 32;
2769 IXGBE_WRITE_REG(hw, IXGBE_VFRE(0), 0);
2770 IXGBE_WRITE_REG(hw, IXGBE_VFRE(1), 0);
2771 IXGBE_WRITE_REG(hw, IXGBE_VFTE(0), 0);
2772 IXGBE_WRITE_REG(hw, IXGBE_VFTE(1), 0);
2773 /* Enable only the PF's pool for Tx/Rx */
2774 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2775 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2776 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2777 ixgbe_set_vmolr(hw, adapter->num_vfs, true);
2778 }
2779
2780 /* Program MRQC for the distribution of queues */
2781 mrqc = ixgbe_setup_mrqc(adapter);
2782
2783 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2784 /* Fill out redirection table */
2785 for (i = 0, j = 0; i < 128; i++, j++) {
2786 if (j == adapter->ring_feature[RING_F_RSS].indices)
2787 j = 0;
2788 /* reta = 4-byte sliding window of
2789 * 0x00..(indices-1)(indices-1)00..etc. */
2790 reta = (reta << 8) | (j * 0x11);
2791 if ((i & 3) == 3)
2792 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2793 }
2794
2795 /* Fill out hash function seeds */
2796 for (i = 0; i < 10; i++)
2797 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2798
2799 if (hw->mac.type == ixgbe_mac_82598EB)
2800 mrqc |= IXGBE_MRQC_RSSEN;
2801 /* Perform hash on these packet types */
2802 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2803 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2804 | IXGBE_MRQC_RSS_FIELD_IPV6
2805 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2806 }
2807 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2808
2809 if (adapter->num_vfs) {
2810 u32 reg;
2811
2812 /* Map PF MAC address in RAR Entry 0 to first pool
2813 * following VFs */
2814 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2815
2816 /* Set up VF register offsets for selected VT Mode, i.e.
2817 * 64 VFs for SR-IOV */
2818 reg = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2819 reg |= IXGBE_GCR_EXT_SRIOV;
2820 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, reg);
2821 }
2822
2823 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2824
2825 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2826 adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2827 /* Disable indicating checksum in descriptor, enables
2828 * RSS hash */
2829 rxcsum |= IXGBE_RXCSUM_PCSD;
2830 }
2831 if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2832 /* Enable IPv4 payload checksum for UDP fragments
2833 * if PCSD is not set */
2834 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2835 }
2836
2837 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2838
2839 if (hw->mac.type == ixgbe_mac_82599EB) {
2840 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2841 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2842 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2843 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2844 }
2845
2846 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2847 /* Enable 82599 HW-RSC */
2848 for (i = 0; i < adapter->num_rx_queues; i++)
2849 ixgbe_configure_rscctl(adapter, i);
2850
2851 /* Disable RSC for ACK packets */
2852 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2853 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2854 }
2855 }
2856
2857 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2858 {
2859 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2860 struct ixgbe_hw *hw = &adapter->hw;
2861 int pool_ndx = adapter->num_vfs;
2862
2863 /* add VID to filter table */
2864 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
2865 }
2866
2867 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2868 {
2869 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2870 struct ixgbe_hw *hw = &adapter->hw;
2871 int pool_ndx = adapter->num_vfs;
2872
2873 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2874 ixgbe_irq_disable(adapter);
2875
2876 vlan_group_set_device(adapter->vlgrp, vid, NULL);
2877
2878 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2879 ixgbe_irq_enable(adapter);
2880
2881 /* remove VID from filter table */
2882 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
2883 }
2884
2885 /**
2886 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
2887 * @adapter: driver data
2888 */
2889 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
2890 {
2891 struct ixgbe_hw *hw = &adapter->hw;
2892 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2893 int i, j;
2894
2895 switch (hw->mac.type) {
2896 case ixgbe_mac_82598EB:
2897 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2898 #ifdef CONFIG_IXGBE_DCB
2899 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
2900 vlnctrl &= ~IXGBE_VLNCTRL_VME;
2901 #endif
2902 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2903 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2904 break;
2905 case ixgbe_mac_82599EB:
2906 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2907 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2908 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2909 #ifdef CONFIG_IXGBE_DCB
2910 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
2911 break;
2912 #endif
2913 for (i = 0; i < adapter->num_rx_queues; i++) {
2914 j = adapter->rx_ring[i]->reg_idx;
2915 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2916 vlnctrl &= ~IXGBE_RXDCTL_VME;
2917 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2918 }
2919 break;
2920 default:
2921 break;
2922 }
2923 }
2924
2925 /**
2926 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
2927 * @adapter: driver data
2928 */
2929 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
2930 {
2931 struct ixgbe_hw *hw = &adapter->hw;
2932 u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2933 int i, j;
2934
2935 switch (hw->mac.type) {
2936 case ixgbe_mac_82598EB:
2937 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2938 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2939 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2940 break;
2941 case ixgbe_mac_82599EB:
2942 vlnctrl |= IXGBE_VLNCTRL_VFE;
2943 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2944 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2945 for (i = 0; i < adapter->num_rx_queues; i++) {
2946 j = adapter->rx_ring[i]->reg_idx;
2947 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2948 vlnctrl |= IXGBE_RXDCTL_VME;
2949 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2950 }
2951 break;
2952 default:
2953 break;
2954 }
2955 }
2956
2957 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2958 struct vlan_group *grp)
2959 {
2960 struct ixgbe_adapter *adapter = netdev_priv(netdev);
2961
2962 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2963 ixgbe_irq_disable(adapter);
2964 adapter->vlgrp = grp;
2965
2966 /*
2967 * For a DCB driver, always enable VLAN tag stripping so we can
2968 * still receive traffic from a DCB-enabled host even if we're
2969 * not in DCB mode.
2970 */
2971 ixgbe_vlan_filter_enable(adapter);
2972
2973 ixgbe_vlan_rx_add_vid(netdev, 0);
2974
2975 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2976 ixgbe_irq_enable(adapter);
2977 }
2978
2979 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2980 {
2981 ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2982
2983 if (adapter->vlgrp) {
2984 u16 vid;
2985 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2986 if (!vlan_group_get_device(adapter->vlgrp, vid))
2987 continue;
2988 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2989 }
2990 }
2991 }
2992
2993 /**
2994 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
2995 * @netdev: network interface device structure
2996 *
2997 * Writes unicast address list to the RAR table.
2998 * Returns: -ENOMEM on failure/insufficient address space
2999 * 0 on no addresses written
3000 * X on writing X addresses to the RAR table
3001 **/
3002 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3003 {
3004 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3005 struct ixgbe_hw *hw = &adapter->hw;
3006 unsigned int vfn = adapter->num_vfs;
3007 unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3008 int count = 0;
3009
3010 /* return ENOMEM indicating insufficient memory for addresses */
3011 if (netdev_uc_count(netdev) > rar_entries)
3012 return -ENOMEM;
3013
3014 if (!netdev_uc_empty(netdev) && rar_entries) {
3015 struct netdev_hw_addr *ha;
3016 /* return error if we do not support writing to RAR table */
3017 if (!hw->mac.ops.set_rar)
3018 return -ENOMEM;
3019
3020 netdev_for_each_uc_addr(ha, netdev) {
3021 if (!rar_entries)
3022 break;
3023 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3024 vfn, IXGBE_RAH_AV);
3025 count++;
3026 }
3027 }
3028 /* write the addresses in reverse order to avoid write combining */
3029 for (; rar_entries > 0 ; rar_entries--)
3030 hw->mac.ops.clear_rar(hw, rar_entries);
3031
3032 return count;
3033 }
3034
3035 /**
3036 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3037 * @netdev: network interface device structure
3038 *
3039 * The set_rx_method entry point is called whenever the unicast/multicast
3040 * address list or the network interface flags are updated. This routine is
3041 * responsible for configuring the hardware for proper unicast, multicast and
3042 * promiscuous mode.
3043 **/
3044 void ixgbe_set_rx_mode(struct net_device *netdev)
3045 {
3046 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3047 struct ixgbe_hw *hw = &adapter->hw;
3048 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3049 int count;
3050
3051 /* Check for Promiscuous and All Multicast modes */
3052
3053 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3054
3055 /* clear the bits we are changing the status of */
3056 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3057
3058 if (netdev->flags & IFF_PROMISC) {
3059 hw->addr_ctrl.user_set_promisc = true;
3060 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3061 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3062 /* don't hardware filter vlans in promisc mode */
3063 ixgbe_vlan_filter_disable(adapter);
3064 } else {
3065 if (netdev->flags & IFF_ALLMULTI) {
3066 fctrl |= IXGBE_FCTRL_MPE;
3067 vmolr |= IXGBE_VMOLR_MPE;
3068 } else {
3069 /*
3070 * Write addresses to the MTA, if the attempt fails
3071 * then we should just turn on promiscous mode so
3072 * that we can at least receive multicast traffic
3073 */
3074 hw->mac.ops.update_mc_addr_list(hw, netdev);
3075 vmolr |= IXGBE_VMOLR_ROMPE;
3076 }
3077 ixgbe_vlan_filter_enable(adapter);
3078 hw->addr_ctrl.user_set_promisc = false;
3079 /*
3080 * Write addresses to available RAR registers, if there is not
3081 * sufficient space to store all the addresses then enable
3082 * unicast promiscous mode
3083 */
3084 count = ixgbe_write_uc_addr_list(netdev);
3085 if (count < 0) {
3086 fctrl |= IXGBE_FCTRL_UPE;
3087 vmolr |= IXGBE_VMOLR_ROPE;
3088 }
3089 }
3090
3091 if (adapter->num_vfs) {
3092 ixgbe_restore_vf_multicasts(adapter);
3093 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3094 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3095 IXGBE_VMOLR_ROPE);
3096 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3097 }
3098
3099 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3100 }
3101
3102 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3103 {
3104 int q_idx;
3105 struct ixgbe_q_vector *q_vector;
3106 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3107
3108 /* legacy and MSI only use one vector */
3109 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3110 q_vectors = 1;
3111
3112 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3113 struct napi_struct *napi;
3114 q_vector = adapter->q_vector[q_idx];
3115 napi = &q_vector->napi;
3116 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3117 if (!q_vector->rxr_count || !q_vector->txr_count) {
3118 if (q_vector->txr_count == 1)
3119 napi->poll = &ixgbe_clean_txonly;
3120 else if (q_vector->rxr_count == 1)
3121 napi->poll = &ixgbe_clean_rxonly;
3122 }
3123 }
3124
3125 napi_enable(napi);
3126 }
3127 }
3128
3129 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3130 {
3131 int q_idx;
3132 struct ixgbe_q_vector *q_vector;
3133 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3134
3135 /* legacy and MSI only use one vector */
3136 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3137 q_vectors = 1;
3138
3139 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3140 q_vector = adapter->q_vector[q_idx];
3141 napi_disable(&q_vector->napi);
3142 }
3143 }
3144
3145 #ifdef CONFIG_IXGBE_DCB
3146 /*
3147 * ixgbe_configure_dcb - Configure DCB hardware
3148 * @adapter: ixgbe adapter struct
3149 *
3150 * This is called by the driver on open to configure the DCB hardware.
3151 * This is also called by the gennetlink interface when reconfiguring
3152 * the DCB state.
3153 */
3154 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3155 {
3156 struct ixgbe_hw *hw = &adapter->hw;
3157 u32 txdctl;
3158 int i, j;
3159
3160 ixgbe_dcb_check_config(&adapter->dcb_cfg);
3161 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3162 ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3163
3164 /* reconfigure the hardware */
3165 ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3166
3167 for (i = 0; i < adapter->num_tx_queues; i++) {
3168 j = adapter->tx_ring[i]->reg_idx;
3169 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3170 /* PThresh workaround for Tx hang with DFP enabled. */
3171 txdctl |= 32;
3172 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3173 }
3174 /* Enable VLAN tag insert/strip */
3175 ixgbe_vlan_filter_enable(adapter);
3176
3177 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3178 }
3179
3180 #endif
3181 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3182 {
3183 struct net_device *netdev = adapter->netdev;
3184 struct ixgbe_hw *hw = &adapter->hw;
3185 int i;
3186
3187 ixgbe_set_rx_mode(netdev);
3188
3189 ixgbe_restore_vlan(adapter);
3190 #ifdef CONFIG_IXGBE_DCB
3191 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3192 if (hw->mac.type == ixgbe_mac_82598EB)
3193 netif_set_gso_max_size(netdev, 32768);
3194 else
3195 netif_set_gso_max_size(netdev, 65536);
3196 ixgbe_configure_dcb(adapter);
3197 } else {
3198 netif_set_gso_max_size(netdev, 65536);
3199 }
3200 #else
3201 netif_set_gso_max_size(netdev, 65536);
3202 #endif
3203
3204 #ifdef IXGBE_FCOE
3205 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3206 ixgbe_configure_fcoe(adapter);
3207
3208 #endif /* IXGBE_FCOE */
3209 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3210 for (i = 0; i < adapter->num_tx_queues; i++)
3211 adapter->tx_ring[i]->atr_sample_rate =
3212 adapter->atr_sample_rate;
3213 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3214 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3215 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3216 }
3217
3218 ixgbe_configure_tx(adapter);
3219 ixgbe_configure_rx(adapter);
3220 for (i = 0; i < adapter->num_rx_queues; i++)
3221 ixgbe_alloc_rx_buffers(adapter, adapter->rx_ring[i],
3222 (adapter->rx_ring[i]->count - 1));
3223 }
3224
3225 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3226 {
3227 switch (hw->phy.type) {
3228 case ixgbe_phy_sfp_avago:
3229 case ixgbe_phy_sfp_ftl:
3230 case ixgbe_phy_sfp_intel:
3231 case ixgbe_phy_sfp_unknown:
3232 case ixgbe_phy_sfp_passive_tyco:
3233 case ixgbe_phy_sfp_passive_unknown:
3234 case ixgbe_phy_sfp_active_unknown:
3235 case ixgbe_phy_sfp_ftl_active:
3236 return true;
3237 default:
3238 return false;
3239 }
3240 }
3241
3242 /**
3243 * ixgbe_sfp_link_config - set up SFP+ link
3244 * @adapter: pointer to private adapter struct
3245 **/
3246 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3247 {
3248 struct ixgbe_hw *hw = &adapter->hw;
3249
3250 if (hw->phy.multispeed_fiber) {
3251 /*
3252 * In multispeed fiber setups, the device may not have
3253 * had a physical connection when the driver loaded.
3254 * If that's the case, the initial link configuration
3255 * couldn't get the MAC into 10G or 1G mode, so we'll
3256 * never have a link status change interrupt fire.
3257 * We need to try and force an autonegotiation
3258 * session, then bring up link.
3259 */
3260 hw->mac.ops.setup_sfp(hw);
3261 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3262 schedule_work(&adapter->multispeed_fiber_task);
3263 } else {
3264 /*
3265 * Direct Attach Cu and non-multispeed fiber modules
3266 * still need to be configured properly prior to
3267 * attempting link.
3268 */
3269 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3270 schedule_work(&adapter->sfp_config_module_task);
3271 }
3272 }
3273
3274 /**
3275 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3276 * @hw: pointer to private hardware struct
3277 *
3278 * Returns 0 on success, negative on failure
3279 **/
3280 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3281 {
3282 u32 autoneg;
3283 bool negotiation, link_up = false;
3284 u32 ret = IXGBE_ERR_LINK_SETUP;
3285
3286 if (hw->mac.ops.check_link)
3287 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3288
3289 if (ret)
3290 goto link_cfg_out;
3291
3292 if (hw->mac.ops.get_link_capabilities)
3293 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
3294 if (ret)
3295 goto link_cfg_out;
3296
3297 if (hw->mac.ops.setup_link)
3298 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3299 link_cfg_out:
3300 return ret;
3301 }
3302
3303 #define IXGBE_MAX_RX_DESC_POLL 10
3304 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3305 int rxr)
3306 {
3307 int j = adapter->rx_ring[rxr]->reg_idx;
3308 int k;
3309
3310 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
3311 if (IXGBE_READ_REG(&adapter->hw,
3312 IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
3313 break;
3314 else
3315 msleep(1);
3316 }
3317 if (k >= IXGBE_MAX_RX_DESC_POLL) {
3318 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3319 "the polling period\n", rxr);
3320 }
3321 ixgbe_release_rx_desc(&adapter->hw, adapter->rx_ring[rxr],
3322 (adapter->rx_ring[rxr]->count - 1));
3323 }
3324
3325 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3326 {
3327 struct net_device *netdev = adapter->netdev;
3328 struct ixgbe_hw *hw = &adapter->hw;
3329 int i, j = 0;
3330 int num_rx_rings = adapter->num_rx_queues;
3331 int err;
3332 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3333 u32 txdctl, rxdctl, mhadd;
3334 u32 dmatxctl;
3335 u32 gpie;
3336 u32 ctrl_ext;
3337
3338 ixgbe_get_hw_control(adapter);
3339
3340 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
3341 (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
3342 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3343 gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
3344 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
3345 } else {
3346 /* MSI only */
3347 gpie = 0;
3348 }
3349 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3350 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3351 gpie |= IXGBE_GPIE_VTMODE_64;
3352 }
3353 /* XXX: to interrupt immediately for EICS writes, enable this */
3354 /* gpie |= IXGBE_GPIE_EIMEN; */
3355 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3356 }
3357
3358 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3359 /*
3360 * use EIAM to auto-mask when MSI-X interrupt is asserted
3361 * this saves a register write for every interrupt
3362 */
3363 switch (hw->mac.type) {
3364 case ixgbe_mac_82598EB:
3365 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3366 break;
3367 default:
3368 case ixgbe_mac_82599EB:
3369 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3370 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3371 break;
3372 }
3373 } else {
3374 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3375 * specifically only auto mask tx and rx interrupts */
3376 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3377 }
3378
3379 /* Enable Thermal over heat sensor interrupt */
3380 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3381 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3382 gpie |= IXGBE_SDP0_GPIEN;
3383 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3384 }
3385
3386 /* Enable fan failure interrupt if media type is copper */
3387 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3388 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3389 gpie |= IXGBE_SDP1_GPIEN;
3390 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3391 }
3392
3393 if (hw->mac.type == ixgbe_mac_82599EB) {
3394 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
3395 gpie |= IXGBE_SDP1_GPIEN;
3396 gpie |= IXGBE_SDP2_GPIEN;
3397 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3398 }
3399
3400 #ifdef IXGBE_FCOE
3401 /* adjust max frame to be able to do baby jumbo for FCoE */
3402 if ((netdev->features & NETIF_F_FCOE_MTU) &&
3403 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3404 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3405
3406 #endif /* IXGBE_FCOE */
3407 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3408 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3409 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3410 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3411
3412 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3413 }
3414
3415 for (i = 0; i < adapter->num_tx_queues; i++) {
3416 j = adapter->tx_ring[i]->reg_idx;
3417 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3418 if (adapter->rx_itr_setting == 0) {
3419 /* cannot set wthresh when itr==0 */
3420 txdctl &= ~0x007F0000;
3421 } else {
3422 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
3423 txdctl |= (8 << 16);
3424 }
3425 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3426 }
3427
3428 if (hw->mac.type == ixgbe_mac_82599EB) {
3429 /* DMATXCTL.EN must be set after all Tx queue config is done */
3430 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3431 dmatxctl |= IXGBE_DMATXCTL_TE;
3432 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3433 }
3434 for (i = 0; i < adapter->num_tx_queues; i++) {
3435 j = adapter->tx_ring[i]->reg_idx;
3436 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3437 txdctl |= IXGBE_TXDCTL_ENABLE;
3438 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3439 if (hw->mac.type == ixgbe_mac_82599EB) {
3440 int wait_loop = 10;
3441 /* poll for Tx Enable ready */
3442 do {
3443 msleep(1);
3444 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3445 } while (--wait_loop &&
3446 !(txdctl & IXGBE_TXDCTL_ENABLE));
3447 if (!wait_loop)
3448 e_err(drv, "Could not enable Tx Queue %d\n", j);
3449 }
3450 }
3451
3452 for (i = 0; i < num_rx_rings; i++) {
3453 j = adapter->rx_ring[i]->reg_idx;
3454 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3455 /* enable PTHRESH=32 descriptors (half the internal cache)
3456 * and HTHRESH=0 descriptors (to minimize latency on fetch),
3457 * this also removes a pesky rx_no_buffer_count increment */
3458 rxdctl |= 0x0020;
3459 rxdctl |= IXGBE_RXDCTL_ENABLE;
3460 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
3461 if (hw->mac.type == ixgbe_mac_82599EB)
3462 ixgbe_rx_desc_queue_enable(adapter, i);
3463 }
3464 /* enable all receives */
3465 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3466 if (hw->mac.type == ixgbe_mac_82598EB)
3467 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
3468 else
3469 rxdctl |= IXGBE_RXCTRL_RXEN;
3470 hw->mac.ops.enable_rx_dma(hw, rxdctl);
3471
3472 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3473 ixgbe_configure_msix(adapter);
3474 else
3475 ixgbe_configure_msi_and_legacy(adapter);
3476
3477 /* enable the optics */
3478 if (hw->phy.multispeed_fiber)
3479 hw->mac.ops.enable_tx_laser(hw);
3480
3481 clear_bit(__IXGBE_DOWN, &adapter->state);
3482 ixgbe_napi_enable_all(adapter);
3483
3484 /* clear any pending interrupts, may auto mask */
3485 IXGBE_READ_REG(hw, IXGBE_EICR);
3486
3487 ixgbe_irq_enable(adapter);
3488
3489 /*
3490 * If this adapter has a fan, check to see if we had a failure
3491 * before we enabled the interrupt.
3492 */
3493 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3494 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3495 if (esdp & IXGBE_ESDP_SDP1)
3496 e_crit(drv, "Fan has stopped, replace the adapter\n");
3497 }
3498
3499 /*
3500 * For hot-pluggable SFP+ devices, a new SFP+ module may have
3501 * arrived before interrupts were enabled but after probe. Such
3502 * devices wouldn't have their type identified yet. We need to
3503 * kick off the SFP+ module setup first, then try to bring up link.
3504 * If we're not hot-pluggable SFP+, we just need to configure link
3505 * and bring it up.
3506 */
3507 if (hw->phy.type == ixgbe_phy_unknown) {
3508 err = hw->phy.ops.identify(hw);
3509 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3510 /*
3511 * Take the device down and schedule the sfp tasklet
3512 * which will unregister_netdev and log it.
3513 */
3514 ixgbe_down(adapter);
3515 schedule_work(&adapter->sfp_config_module_task);
3516 return err;
3517 }
3518 }
3519
3520 if (ixgbe_is_sfp(hw)) {
3521 ixgbe_sfp_link_config(adapter);
3522 } else {
3523 err = ixgbe_non_sfp_link_config(hw);
3524 if (err)
3525 e_err(probe, "link_config FAILED %d\n", err);
3526 }
3527
3528 for (i = 0; i < adapter->num_tx_queues; i++)
3529 set_bit(__IXGBE_FDIR_INIT_DONE,
3530 &(adapter->tx_ring[i]->reinit_state));
3531
3532 /* enable transmits */
3533 netif_tx_start_all_queues(netdev);
3534
3535 /* bring the link up in the watchdog, this could race with our first
3536 * link up interrupt but shouldn't be a problem */
3537 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3538 adapter->link_check_timeout = jiffies;
3539 mod_timer(&adapter->watchdog_timer, jiffies);
3540
3541 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3542 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3543 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3544 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3545
3546 return 0;
3547 }
3548
3549 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3550 {
3551 WARN_ON(in_interrupt());
3552 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3553 msleep(1);
3554 ixgbe_down(adapter);
3555 /*
3556 * If SR-IOV enabled then wait a bit before bringing the adapter
3557 * back up to give the VFs time to respond to the reset. The
3558 * two second wait is based upon the watchdog timer cycle in
3559 * the VF driver.
3560 */
3561 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3562 msleep(2000);
3563 ixgbe_up(adapter);
3564 clear_bit(__IXGBE_RESETTING, &adapter->state);
3565 }
3566
3567 int ixgbe_up(struct ixgbe_adapter *adapter)
3568 {
3569 /* hardware has been reset, we need to reload some things */
3570 ixgbe_configure(adapter);
3571
3572 return ixgbe_up_complete(adapter);
3573 }
3574
3575 void ixgbe_reset(struct ixgbe_adapter *adapter)
3576 {
3577 struct ixgbe_hw *hw = &adapter->hw;
3578 int err;
3579
3580 err = hw->mac.ops.init_hw(hw);
3581 switch (err) {
3582 case 0:
3583 case IXGBE_ERR_SFP_NOT_PRESENT:
3584 break;
3585 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3586 e_dev_err("master disable timed out\n");
3587 break;
3588 case IXGBE_ERR_EEPROM_VERSION:
3589 /* We are running on a pre-production device, log a warning */
3590 e_dev_warn("This device is a pre-production adapter/LOM. "
3591 "Please be aware there may be issuesassociated with "
3592 "your hardware. If you are experiencing problems "
3593 "please contact your Intel or hardware "
3594 "representative who provided you with this "
3595 "hardware.\n");
3596 break;
3597 default:
3598 e_dev_err("Hardware Error: %d\n", err);
3599 }
3600
3601 /* reprogram the RAR[0] in case user changed it. */
3602 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3603 IXGBE_RAH_AV);
3604 }
3605
3606 /**
3607 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3608 * @adapter: board private structure
3609 * @rx_ring: ring to free buffers from
3610 **/
3611 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3612 struct ixgbe_ring *rx_ring)
3613 {
3614 struct pci_dev *pdev = adapter->pdev;
3615 unsigned long size;
3616 unsigned int i;
3617
3618 /* Free all the Rx ring sk_buffs */
3619
3620 for (i = 0; i < rx_ring->count; i++) {
3621 struct ixgbe_rx_buffer *rx_buffer_info;
3622
3623 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3624 if (rx_buffer_info->dma) {
3625 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3626 rx_ring->rx_buf_len,
3627 DMA_FROM_DEVICE);
3628 rx_buffer_info->dma = 0;
3629 }
3630 if (rx_buffer_info->skb) {
3631 struct sk_buff *skb = rx_buffer_info->skb;
3632 rx_buffer_info->skb = NULL;
3633 do {
3634 struct sk_buff *this = skb;
3635 if (IXGBE_RSC_CB(this)->delay_unmap) {
3636 dma_unmap_single(&pdev->dev,
3637 IXGBE_RSC_CB(this)->dma,
3638 rx_ring->rx_buf_len,
3639 DMA_FROM_DEVICE);
3640 IXGBE_RSC_CB(this)->dma = 0;
3641 IXGBE_RSC_CB(skb)->delay_unmap = false;
3642 }
3643 skb = skb->prev;
3644 dev_kfree_skb(this);
3645 } while (skb);
3646 }
3647 if (!rx_buffer_info->page)
3648 continue;
3649 if (rx_buffer_info->page_dma) {
3650 dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3651 PAGE_SIZE / 2, DMA_FROM_DEVICE);
3652 rx_buffer_info->page_dma = 0;
3653 }
3654 put_page(rx_buffer_info->page);
3655 rx_buffer_info->page = NULL;
3656 rx_buffer_info->page_offset = 0;
3657 }
3658
3659 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3660 memset(rx_ring->rx_buffer_info, 0, size);
3661
3662 /* Zero out the descriptor ring */
3663 memset(rx_ring->desc, 0, rx_ring->size);
3664
3665 rx_ring->next_to_clean = 0;
3666 rx_ring->next_to_use = 0;
3667
3668 if (rx_ring->head)
3669 writel(0, adapter->hw.hw_addr + rx_ring->head);
3670 if (rx_ring->tail)
3671 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3672 }
3673
3674 /**
3675 * ixgbe_clean_tx_ring - Free Tx Buffers
3676 * @adapter: board private structure
3677 * @tx_ring: ring to be cleaned
3678 **/
3679 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3680 struct ixgbe_ring *tx_ring)
3681 {
3682 struct ixgbe_tx_buffer *tx_buffer_info;
3683 unsigned long size;
3684 unsigned int i;
3685
3686 /* Free all the Tx ring sk_buffs */
3687
3688 for (i = 0; i < tx_ring->count; i++) {
3689 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3690 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3691 }
3692
3693 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3694 memset(tx_ring->tx_buffer_info, 0, size);
3695
3696 /* Zero out the descriptor ring */
3697 memset(tx_ring->desc, 0, tx_ring->size);
3698
3699 tx_ring->next_to_use = 0;
3700 tx_ring->next_to_clean = 0;
3701
3702 if (tx_ring->head)
3703 writel(0, adapter->hw.hw_addr + tx_ring->head);
3704 if (tx_ring->tail)
3705 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3706 }
3707
3708 /**
3709 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3710 * @adapter: board private structure
3711 **/
3712 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3713 {
3714 int i;
3715
3716 for (i = 0; i < adapter->num_rx_queues; i++)
3717 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3718 }
3719
3720 /**
3721 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3722 * @adapter: board private structure
3723 **/
3724 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3725 {
3726 int i;
3727
3728 for (i = 0; i < adapter->num_tx_queues; i++)
3729 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3730 }
3731
3732 void ixgbe_down(struct ixgbe_adapter *adapter)
3733 {
3734 struct net_device *netdev = adapter->netdev;
3735 struct ixgbe_hw *hw = &adapter->hw;
3736 u32 rxctrl;
3737 u32 txdctl;
3738 int i, j;
3739
3740 /* signal that we are down to the interrupt handler */
3741 set_bit(__IXGBE_DOWN, &adapter->state);
3742
3743 /* disable receive for all VFs and wait one second */
3744 if (adapter->num_vfs) {
3745 /* ping all the active vfs to let them know we are going down */
3746 ixgbe_ping_all_vfs(adapter);
3747
3748 /* Disable all VFTE/VFRE TX/RX */
3749 ixgbe_disable_tx_rx(adapter);
3750
3751 /* Mark all the VFs as inactive */
3752 for (i = 0 ; i < adapter->num_vfs; i++)
3753 adapter->vfinfo[i].clear_to_send = 0;
3754 }
3755
3756 /* disable receives */
3757 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3758 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3759
3760 IXGBE_WRITE_FLUSH(hw);
3761 msleep(10);
3762
3763 netif_tx_stop_all_queues(netdev);
3764
3765 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3766 del_timer_sync(&adapter->sfp_timer);
3767 del_timer_sync(&adapter->watchdog_timer);
3768 cancel_work_sync(&adapter->watchdog_task);
3769
3770 netif_carrier_off(netdev);
3771 netif_tx_disable(netdev);
3772
3773 ixgbe_irq_disable(adapter);
3774
3775 ixgbe_napi_disable_all(adapter);
3776
3777 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3778 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3779 cancel_work_sync(&adapter->fdir_reinit_task);
3780
3781 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3782 cancel_work_sync(&adapter->check_overtemp_task);
3783
3784 /* disable transmits in the hardware now that interrupts are off */
3785 for (i = 0; i < adapter->num_tx_queues; i++) {
3786 j = adapter->tx_ring[i]->reg_idx;
3787 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3788 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3789 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3790 }
3791 /* Disable the Tx DMA engine on 82599 */
3792 if (hw->mac.type == ixgbe_mac_82599EB)
3793 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3794 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3795 ~IXGBE_DMATXCTL_TE));
3796
3797 /* power down the optics */
3798 if (hw->phy.multispeed_fiber)
3799 hw->mac.ops.disable_tx_laser(hw);
3800
3801 /* clear n-tuple filters that are cached */
3802 ethtool_ntuple_flush(netdev);
3803
3804 if (!pci_channel_offline(adapter->pdev))
3805 ixgbe_reset(adapter);
3806 ixgbe_clean_all_tx_rings(adapter);
3807 ixgbe_clean_all_rx_rings(adapter);
3808
3809 #ifdef CONFIG_IXGBE_DCA
3810 /* since we reset the hardware DCA settings were cleared */
3811 ixgbe_setup_dca(adapter);
3812 #endif
3813 }
3814
3815 /**
3816 * ixgbe_poll - NAPI Rx polling callback
3817 * @napi: structure for representing this polling device
3818 * @budget: how many packets driver is allowed to clean
3819 *
3820 * This function is used for legacy and MSI, NAPI mode
3821 **/
3822 static int ixgbe_poll(struct napi_struct *napi, int budget)
3823 {
3824 struct ixgbe_q_vector *q_vector =
3825 container_of(napi, struct ixgbe_q_vector, napi);
3826 struct ixgbe_adapter *adapter = q_vector->adapter;
3827 int tx_clean_complete, work_done = 0;
3828
3829 #ifdef CONFIG_IXGBE_DCA
3830 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3831 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3832 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3833 }
3834 #endif
3835
3836 tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3837 ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3838
3839 if (!tx_clean_complete)
3840 work_done = budget;
3841
3842 /* If budget not fully consumed, exit the polling mode */
3843 if (work_done < budget) {
3844 napi_complete(napi);
3845 if (adapter->rx_itr_setting & 1)
3846 ixgbe_set_itr(adapter);
3847 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3848 ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3849 }
3850 return work_done;
3851 }
3852
3853 /**
3854 * ixgbe_tx_timeout - Respond to a Tx Hang
3855 * @netdev: network interface device structure
3856 **/
3857 static void ixgbe_tx_timeout(struct net_device *netdev)
3858 {
3859 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3860
3861 /* Do the reset outside of interrupt context */
3862 schedule_work(&adapter->reset_task);
3863 }
3864
3865 static void ixgbe_reset_task(struct work_struct *work)
3866 {
3867 struct ixgbe_adapter *adapter;
3868 adapter = container_of(work, struct ixgbe_adapter, reset_task);
3869
3870 /* If we're already down or resetting, just bail */
3871 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3872 test_bit(__IXGBE_RESETTING, &adapter->state))
3873 return;
3874
3875 adapter->tx_timeout_count++;
3876
3877 ixgbe_dump(adapter);
3878 netdev_err(adapter->netdev, "Reset adapter\n");
3879 ixgbe_reinit_locked(adapter);
3880 }
3881
3882 #ifdef CONFIG_IXGBE_DCB
3883 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3884 {
3885 bool ret = false;
3886 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3887
3888 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3889 return ret;
3890
3891 f->mask = 0x7 << 3;
3892 adapter->num_rx_queues = f->indices;
3893 adapter->num_tx_queues = f->indices;
3894 ret = true;
3895
3896 return ret;
3897 }
3898 #endif
3899
3900 /**
3901 * ixgbe_set_rss_queues: Allocate queues for RSS
3902 * @adapter: board private structure to initialize
3903 *
3904 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3905 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3906 *
3907 **/
3908 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3909 {
3910 bool ret = false;
3911 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3912
3913 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3914 f->mask = 0xF;
3915 adapter->num_rx_queues = f->indices;
3916 adapter->num_tx_queues = f->indices;
3917 ret = true;
3918 } else {
3919 ret = false;
3920 }
3921
3922 return ret;
3923 }
3924
3925 /**
3926 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3927 * @adapter: board private structure to initialize
3928 *
3929 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3930 * to the original CPU that initiated the Tx session. This runs in addition
3931 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3932 * Rx load across CPUs using RSS.
3933 *
3934 **/
3935 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3936 {
3937 bool ret = false;
3938 struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3939
3940 f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3941 f_fdir->mask = 0;
3942
3943 /* Flow Director must have RSS enabled */
3944 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3945 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3946 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3947 adapter->num_tx_queues = f_fdir->indices;
3948 adapter->num_rx_queues = f_fdir->indices;
3949 ret = true;
3950 } else {
3951 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3952 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3953 }
3954 return ret;
3955 }
3956
3957 #ifdef IXGBE_FCOE
3958 /**
3959 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3960 * @adapter: board private structure to initialize
3961 *
3962 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3963 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3964 * rx queues out of the max number of rx queues, instead, it is used as the
3965 * index of the first rx queue used by FCoE.
3966 *
3967 **/
3968 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3969 {
3970 bool ret = false;
3971 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3972
3973 f->indices = min((int)num_online_cpus(), f->indices);
3974 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3975 adapter->num_rx_queues = 1;
3976 adapter->num_tx_queues = 1;
3977 #ifdef CONFIG_IXGBE_DCB
3978 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3979 e_info(probe, "FCoE enabled with DCB\n");
3980 ixgbe_set_dcb_queues(adapter);
3981 }
3982 #endif
3983 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3984 e_info(probe, "FCoE enabled with RSS\n");
3985 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3986 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3987 ixgbe_set_fdir_queues(adapter);
3988 else
3989 ixgbe_set_rss_queues(adapter);
3990 }
3991 /* adding FCoE rx rings to the end */
3992 f->mask = adapter->num_rx_queues;
3993 adapter->num_rx_queues += f->indices;
3994 adapter->num_tx_queues += f->indices;
3995
3996 ret = true;
3997 }
3998
3999 return ret;
4000 }
4001
4002 #endif /* IXGBE_FCOE */
4003 /**
4004 * ixgbe_set_sriov_queues: Allocate queues for IOV use
4005 * @adapter: board private structure to initialize
4006 *
4007 * IOV doesn't actually use anything, so just NAK the
4008 * request for now and let the other queue routines
4009 * figure out what to do.
4010 */
4011 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4012 {
4013 return false;
4014 }
4015
4016 /*
4017 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4018 * @adapter: board private structure to initialize
4019 *
4020 * This is the top level queue allocation routine. The order here is very
4021 * important, starting with the "most" number of features turned on at once,
4022 * and ending with the smallest set of features. This way large combinations
4023 * can be allocated if they're turned on, and smaller combinations are the
4024 * fallthrough conditions.
4025 *
4026 **/
4027 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4028 {
4029 /* Start with base case */
4030 adapter->num_rx_queues = 1;
4031 adapter->num_tx_queues = 1;
4032 adapter->num_rx_pools = adapter->num_rx_queues;
4033 adapter->num_rx_queues_per_pool = 1;
4034
4035 if (ixgbe_set_sriov_queues(adapter))
4036 return;
4037
4038 #ifdef IXGBE_FCOE
4039 if (ixgbe_set_fcoe_queues(adapter))
4040 goto done;
4041
4042 #endif /* IXGBE_FCOE */
4043 #ifdef CONFIG_IXGBE_DCB
4044 if (ixgbe_set_dcb_queues(adapter))
4045 goto done;
4046
4047 #endif
4048 if (ixgbe_set_fdir_queues(adapter))
4049 goto done;
4050
4051 if (ixgbe_set_rss_queues(adapter))
4052 goto done;
4053
4054 /* fallback to base case */
4055 adapter->num_rx_queues = 1;
4056 adapter->num_tx_queues = 1;
4057
4058 done:
4059 /* Notify the stack of the (possibly) reduced Tx Queue count. */
4060 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4061 }
4062
4063 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4064 int vectors)
4065 {
4066 int err, vector_threshold;
4067
4068 /* We'll want at least 3 (vector_threshold):
4069 * 1) TxQ[0] Cleanup
4070 * 2) RxQ[0] Cleanup
4071 * 3) Other (Link Status Change, etc.)
4072 * 4) TCP Timer (optional)
4073 */
4074 vector_threshold = MIN_MSIX_COUNT;
4075
4076 /* The more we get, the more we will assign to Tx/Rx Cleanup
4077 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4078 * Right now, we simply care about how many we'll get; we'll
4079 * set them up later while requesting irq's.
4080 */
4081 while (vectors >= vector_threshold) {
4082 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4083 vectors);
4084 if (!err) /* Success in acquiring all requested vectors. */
4085 break;
4086 else if (err < 0)
4087 vectors = 0; /* Nasty failure, quit now */
4088 else /* err == number of vectors we should try again with */
4089 vectors = err;
4090 }
4091
4092 if (vectors < vector_threshold) {
4093 /* Can't allocate enough MSI-X interrupts? Oh well.
4094 * This just means we'll go with either a single MSI
4095 * vector or fall back to legacy interrupts.
4096 */
4097 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4098 "Unable to allocate MSI-X interrupts\n");
4099 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4100 kfree(adapter->msix_entries);
4101 adapter->msix_entries = NULL;
4102 } else {
4103 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4104 /*
4105 * Adjust for only the vectors we'll use, which is minimum
4106 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4107 * vectors we were allocated.
4108 */
4109 adapter->num_msix_vectors = min(vectors,
4110 adapter->max_msix_q_vectors + NON_Q_VECTORS);
4111 }
4112 }
4113
4114 /**
4115 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4116 * @adapter: board private structure to initialize
4117 *
4118 * Cache the descriptor ring offsets for RSS to the assigned rings.
4119 *
4120 **/
4121 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4122 {
4123 int i;
4124 bool ret = false;
4125
4126 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4127 for (i = 0; i < adapter->num_rx_queues; i++)
4128 adapter->rx_ring[i]->reg_idx = i;
4129 for (i = 0; i < adapter->num_tx_queues; i++)
4130 adapter->tx_ring[i]->reg_idx = i;
4131 ret = true;
4132 } else {
4133 ret = false;
4134 }
4135
4136 return ret;
4137 }
4138
4139 #ifdef CONFIG_IXGBE_DCB
4140 /**
4141 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4142 * @adapter: board private structure to initialize
4143 *
4144 * Cache the descriptor ring offsets for DCB to the assigned rings.
4145 *
4146 **/
4147 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4148 {
4149 int i;
4150 bool ret = false;
4151 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4152
4153 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4154 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4155 /* the number of queues is assumed to be symmetric */
4156 for (i = 0; i < dcb_i; i++) {
4157 adapter->rx_ring[i]->reg_idx = i << 3;
4158 adapter->tx_ring[i]->reg_idx = i << 2;
4159 }
4160 ret = true;
4161 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4162 if (dcb_i == 8) {
4163 /*
4164 * Tx TC0 starts at: descriptor queue 0
4165 * Tx TC1 starts at: descriptor queue 32
4166 * Tx TC2 starts at: descriptor queue 64
4167 * Tx TC3 starts at: descriptor queue 80
4168 * Tx TC4 starts at: descriptor queue 96
4169 * Tx TC5 starts at: descriptor queue 104
4170 * Tx TC6 starts at: descriptor queue 112
4171 * Tx TC7 starts at: descriptor queue 120
4172 *
4173 * Rx TC0-TC7 are offset by 16 queues each
4174 */
4175 for (i = 0; i < 3; i++) {
4176 adapter->tx_ring[i]->reg_idx = i << 5;
4177 adapter->rx_ring[i]->reg_idx = i << 4;
4178 }
4179 for ( ; i < 5; i++) {
4180 adapter->tx_ring[i]->reg_idx =
4181 ((i + 2) << 4);
4182 adapter->rx_ring[i]->reg_idx = i << 4;
4183 }
4184 for ( ; i < dcb_i; i++) {
4185 adapter->tx_ring[i]->reg_idx =
4186 ((i + 8) << 3);
4187 adapter->rx_ring[i]->reg_idx = i << 4;
4188 }
4189
4190 ret = true;
4191 } else if (dcb_i == 4) {
4192 /*
4193 * Tx TC0 starts at: descriptor queue 0
4194 * Tx TC1 starts at: descriptor queue 64
4195 * Tx TC2 starts at: descriptor queue 96
4196 * Tx TC3 starts at: descriptor queue 112
4197 *
4198 * Rx TC0-TC3 are offset by 32 queues each
4199 */
4200 adapter->tx_ring[0]->reg_idx = 0;
4201 adapter->tx_ring[1]->reg_idx = 64;
4202 adapter->tx_ring[2]->reg_idx = 96;
4203 adapter->tx_ring[3]->reg_idx = 112;
4204 for (i = 0 ; i < dcb_i; i++)
4205 adapter->rx_ring[i]->reg_idx = i << 5;
4206
4207 ret = true;
4208 } else {
4209 ret = false;
4210 }
4211 } else {
4212 ret = false;
4213 }
4214 } else {
4215 ret = false;
4216 }
4217
4218 return ret;
4219 }
4220 #endif
4221
4222 /**
4223 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4224 * @adapter: board private structure to initialize
4225 *
4226 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4227 *
4228 **/
4229 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4230 {
4231 int i;
4232 bool ret = false;
4233
4234 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4235 ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4236 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4237 for (i = 0; i < adapter->num_rx_queues; i++)
4238 adapter->rx_ring[i]->reg_idx = i;
4239 for (i = 0; i < adapter->num_tx_queues; i++)
4240 adapter->tx_ring[i]->reg_idx = i;
4241 ret = true;
4242 }
4243
4244 return ret;
4245 }
4246
4247 #ifdef IXGBE_FCOE
4248 /**
4249 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4250 * @adapter: board private structure to initialize
4251 *
4252 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4253 *
4254 */
4255 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4256 {
4257 int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4258 bool ret = false;
4259 struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4260
4261 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4262 #ifdef CONFIG_IXGBE_DCB
4263 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4264 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4265
4266 ixgbe_cache_ring_dcb(adapter);
4267 /* find out queues in TC for FCoE */
4268 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4269 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4270 /*
4271 * In 82599, the number of Tx queues for each traffic
4272 * class for both 8-TC and 4-TC modes are:
4273 * TCs : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4274 * 8 TCs: 32 32 16 16 8 8 8 8
4275 * 4 TCs: 64 64 32 32
4276 * We have max 8 queues for FCoE, where 8 the is
4277 * FCoE redirection table size. If TC for FCoE is
4278 * less than or equal to TC3, we have enough queues
4279 * to add max of 8 queues for FCoE, so we start FCoE
4280 * tx descriptor from the next one, i.e., reg_idx + 1.
4281 * If TC for FCoE is above TC3, implying 8 TC mode,
4282 * and we need 8 for FCoE, we have to take all queues
4283 * in that traffic class for FCoE.
4284 */
4285 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4286 fcoe_tx_i--;
4287 }
4288 #endif /* CONFIG_IXGBE_DCB */
4289 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4290 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4291 (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4292 ixgbe_cache_ring_fdir(adapter);
4293 else
4294 ixgbe_cache_ring_rss(adapter);
4295
4296 fcoe_rx_i = f->mask;
4297 fcoe_tx_i = f->mask;
4298 }
4299 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4300 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4301 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4302 }
4303 ret = true;
4304 }
4305 return ret;
4306 }
4307
4308 #endif /* IXGBE_FCOE */
4309 /**
4310 * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4311 * @adapter: board private structure to initialize
4312 *
4313 * SR-IOV doesn't use any descriptor rings but changes the default if
4314 * no other mapping is used.
4315 *
4316 */
4317 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4318 {
4319 adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4320 adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4321 if (adapter->num_vfs)
4322 return true;
4323 else
4324 return false;
4325 }
4326
4327 /**
4328 * ixgbe_cache_ring_register - Descriptor ring to register mapping
4329 * @adapter: board private structure to initialize
4330 *
4331 * Once we know the feature-set enabled for the device, we'll cache
4332 * the register offset the descriptor ring is assigned to.
4333 *
4334 * Note, the order the various feature calls is important. It must start with
4335 * the "most" features enabled at the same time, then trickle down to the
4336 * least amount of features turned on at once.
4337 **/
4338 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4339 {
4340 /* start with default case */
4341 adapter->rx_ring[0]->reg_idx = 0;
4342 adapter->tx_ring[0]->reg_idx = 0;
4343
4344 if (ixgbe_cache_ring_sriov(adapter))
4345 return;
4346
4347 #ifdef IXGBE_FCOE
4348 if (ixgbe_cache_ring_fcoe(adapter))
4349 return;
4350
4351 #endif /* IXGBE_FCOE */
4352 #ifdef CONFIG_IXGBE_DCB
4353 if (ixgbe_cache_ring_dcb(adapter))
4354 return;
4355
4356 #endif
4357 if (ixgbe_cache_ring_fdir(adapter))
4358 return;
4359
4360 if (ixgbe_cache_ring_rss(adapter))
4361 return;
4362 }
4363
4364 /**
4365 * ixgbe_alloc_queues - Allocate memory for all rings
4366 * @adapter: board private structure to initialize
4367 *
4368 * We allocate one ring per queue at run-time since we don't know the
4369 * number of queues at compile-time. The polling_netdev array is
4370 * intended for Multiqueue, but should work fine with a single queue.
4371 **/
4372 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4373 {
4374 int i;
4375 int orig_node = adapter->node;
4376
4377 for (i = 0; i < adapter->num_tx_queues; i++) {
4378 struct ixgbe_ring *ring = adapter->tx_ring[i];
4379 if (orig_node == -1) {
4380 int cur_node = next_online_node(adapter->node);
4381 if (cur_node == MAX_NUMNODES)
4382 cur_node = first_online_node;
4383 adapter->node = cur_node;
4384 }
4385 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4386 adapter->node);
4387 if (!ring)
4388 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4389 if (!ring)
4390 goto err_tx_ring_allocation;
4391 ring->count = adapter->tx_ring_count;
4392 ring->queue_index = i;
4393 ring->numa_node = adapter->node;
4394
4395 adapter->tx_ring[i] = ring;
4396 }
4397
4398 /* Restore the adapter's original node */
4399 adapter->node = orig_node;
4400
4401 for (i = 0; i < adapter->num_rx_queues; i++) {
4402 struct ixgbe_ring *ring = adapter->rx_ring[i];
4403 if (orig_node == -1) {
4404 int cur_node = next_online_node(adapter->node);
4405 if (cur_node == MAX_NUMNODES)
4406 cur_node = first_online_node;
4407 adapter->node = cur_node;
4408 }
4409 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4410 adapter->node);
4411 if (!ring)
4412 ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4413 if (!ring)
4414 goto err_rx_ring_allocation;
4415 ring->count = adapter->rx_ring_count;
4416 ring->queue_index = i;
4417 ring->numa_node = adapter->node;
4418
4419 adapter->rx_ring[i] = ring;
4420 }
4421
4422 /* Restore the adapter's original node */
4423 adapter->node = orig_node;
4424
4425 ixgbe_cache_ring_register(adapter);
4426
4427 return 0;
4428
4429 err_rx_ring_allocation:
4430 for (i = 0; i < adapter->num_tx_queues; i++)
4431 kfree(adapter->tx_ring[i]);
4432 err_tx_ring_allocation:
4433 return -ENOMEM;
4434 }
4435
4436 /**
4437 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4438 * @adapter: board private structure to initialize
4439 *
4440 * Attempt to configure the interrupts using the best available
4441 * capabilities of the hardware and the kernel.
4442 **/
4443 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4444 {
4445 struct ixgbe_hw *hw = &adapter->hw;
4446 int err = 0;
4447 int vector, v_budget;
4448
4449 /*
4450 * It's easy to be greedy for MSI-X vectors, but it really
4451 * doesn't do us much good if we have a lot more vectors
4452 * than CPU's. So let's be conservative and only ask for
4453 * (roughly) the same number of vectors as there are CPU's.
4454 */
4455 v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4456 (int)num_online_cpus()) + NON_Q_VECTORS;
4457
4458 /*
4459 * At the same time, hardware can only support a maximum of
4460 * hw.mac->max_msix_vectors vectors. With features
4461 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4462 * descriptor queues supported by our device. Thus, we cap it off in
4463 * those rare cases where the cpu count also exceeds our vector limit.
4464 */
4465 v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4466
4467 /* A failure in MSI-X entry allocation isn't fatal, but it does
4468 * mean we disable MSI-X capabilities of the adapter. */
4469 adapter->msix_entries = kcalloc(v_budget,
4470 sizeof(struct msix_entry), GFP_KERNEL);
4471 if (adapter->msix_entries) {
4472 for (vector = 0; vector < v_budget; vector++)
4473 adapter->msix_entries[vector].entry = vector;
4474
4475 ixgbe_acquire_msix_vectors(adapter, v_budget);
4476
4477 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4478 goto out;
4479 }
4480
4481 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4482 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4483 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4484 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4485 adapter->atr_sample_rate = 0;
4486 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4487 ixgbe_disable_sriov(adapter);
4488
4489 ixgbe_set_num_queues(adapter);
4490
4491 err = pci_enable_msi(adapter->pdev);
4492 if (!err) {
4493 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4494 } else {
4495 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4496 "Unable to allocate MSI interrupt, "
4497 "falling back to legacy. Error: %d\n", err);
4498 /* reset err */
4499 err = 0;
4500 }
4501
4502 out:
4503 return err;
4504 }
4505
4506 /**
4507 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4508 * @adapter: board private structure to initialize
4509 *
4510 * We allocate one q_vector per queue interrupt. If allocation fails we
4511 * return -ENOMEM.
4512 **/
4513 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4514 {
4515 int q_idx, num_q_vectors;
4516 struct ixgbe_q_vector *q_vector;
4517 int napi_vectors;
4518 int (*poll)(struct napi_struct *, int);
4519
4520 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4521 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4522 napi_vectors = adapter->num_rx_queues;
4523 poll = &ixgbe_clean_rxtx_many;
4524 } else {
4525 num_q_vectors = 1;
4526 napi_vectors = 1;
4527 poll = &ixgbe_poll;
4528 }
4529
4530 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4531 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4532 GFP_KERNEL, adapter->node);
4533 if (!q_vector)
4534 q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4535 GFP_KERNEL);
4536 if (!q_vector)
4537 goto err_out;
4538 q_vector->adapter = adapter;
4539 if (q_vector->txr_count && !q_vector->rxr_count)
4540 q_vector->eitr = adapter->tx_eitr_param;
4541 else
4542 q_vector->eitr = adapter->rx_eitr_param;
4543 q_vector->v_idx = q_idx;
4544 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4545 adapter->q_vector[q_idx] = q_vector;
4546 }
4547
4548 return 0;
4549
4550 err_out:
4551 while (q_idx) {
4552 q_idx--;
4553 q_vector = adapter->q_vector[q_idx];
4554 netif_napi_del(&q_vector->napi);
4555 kfree(q_vector);
4556 adapter->q_vector[q_idx] = NULL;
4557 }
4558 return -ENOMEM;
4559 }
4560
4561 /**
4562 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4563 * @adapter: board private structure to initialize
4564 *
4565 * This function frees the memory allocated to the q_vectors. In addition if
4566 * NAPI is enabled it will delete any references to the NAPI struct prior
4567 * to freeing the q_vector.
4568 **/
4569 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4570 {
4571 int q_idx, num_q_vectors;
4572
4573 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4574 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4575 else
4576 num_q_vectors = 1;
4577
4578 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4579 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4580 adapter->q_vector[q_idx] = NULL;
4581 netif_napi_del(&q_vector->napi);
4582 kfree(q_vector);
4583 }
4584 }
4585
4586 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4587 {
4588 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4589 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4590 pci_disable_msix(adapter->pdev);
4591 kfree(adapter->msix_entries);
4592 adapter->msix_entries = NULL;
4593 } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4594 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4595 pci_disable_msi(adapter->pdev);
4596 }
4597 }
4598
4599 /**
4600 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4601 * @adapter: board private structure to initialize
4602 *
4603 * We determine which interrupt scheme to use based on...
4604 * - Kernel support (MSI, MSI-X)
4605 * - which can be user-defined (via MODULE_PARAM)
4606 * - Hardware queue count (num_*_queues)
4607 * - defined by miscellaneous hardware support/features (RSS, etc.)
4608 **/
4609 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4610 {
4611 int err;
4612
4613 /* Number of supported queues */
4614 ixgbe_set_num_queues(adapter);
4615
4616 err = ixgbe_set_interrupt_capability(adapter);
4617 if (err) {
4618 e_dev_err("Unable to setup interrupt capabilities\n");
4619 goto err_set_interrupt;
4620 }
4621
4622 err = ixgbe_alloc_q_vectors(adapter);
4623 if (err) {
4624 e_dev_err("Unable to allocate memory for queue vectors\n");
4625 goto err_alloc_q_vectors;
4626 }
4627
4628 err = ixgbe_alloc_queues(adapter);
4629 if (err) {
4630 e_dev_err("Unable to allocate memory for queues\n");
4631 goto err_alloc_queues;
4632 }
4633
4634 e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4635 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4636 adapter->num_rx_queues, adapter->num_tx_queues);
4637
4638 set_bit(__IXGBE_DOWN, &adapter->state);
4639
4640 return 0;
4641
4642 err_alloc_queues:
4643 ixgbe_free_q_vectors(adapter);
4644 err_alloc_q_vectors:
4645 ixgbe_reset_interrupt_capability(adapter);
4646 err_set_interrupt:
4647 return err;
4648 }
4649
4650 /**
4651 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4652 * @adapter: board private structure to clear interrupt scheme on
4653 *
4654 * We go through and clear interrupt specific resources and reset the structure
4655 * to pre-load conditions
4656 **/
4657 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4658 {
4659 int i;
4660
4661 for (i = 0; i < adapter->num_tx_queues; i++) {
4662 kfree(adapter->tx_ring[i]);
4663 adapter->tx_ring[i] = NULL;
4664 }
4665 for (i = 0; i < adapter->num_rx_queues; i++) {
4666 kfree(adapter->rx_ring[i]);
4667 adapter->rx_ring[i] = NULL;
4668 }
4669
4670 ixgbe_free_q_vectors(adapter);
4671 ixgbe_reset_interrupt_capability(adapter);
4672 }
4673
4674 /**
4675 * ixgbe_sfp_timer - worker thread to find a missing module
4676 * @data: pointer to our adapter struct
4677 **/
4678 static void ixgbe_sfp_timer(unsigned long data)
4679 {
4680 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4681
4682 /*
4683 * Do the sfp_timer outside of interrupt context due to the
4684 * delays that sfp+ detection requires
4685 */
4686 schedule_work(&adapter->sfp_task);
4687 }
4688
4689 /**
4690 * ixgbe_sfp_task - worker thread to find a missing module
4691 * @work: pointer to work_struct containing our data
4692 **/
4693 static void ixgbe_sfp_task(struct work_struct *work)
4694 {
4695 struct ixgbe_adapter *adapter = container_of(work,
4696 struct ixgbe_adapter,
4697 sfp_task);
4698 struct ixgbe_hw *hw = &adapter->hw;
4699
4700 if ((hw->phy.type == ixgbe_phy_nl) &&
4701 (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4702 s32 ret = hw->phy.ops.identify_sfp(hw);
4703 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4704 goto reschedule;
4705 ret = hw->phy.ops.reset(hw);
4706 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4707 e_dev_err("failed to initialize because an unsupported "
4708 "SFP+ module type was detected.\n");
4709 e_dev_err("Reload the driver after installing a "
4710 "supported module.\n");
4711 unregister_netdev(adapter->netdev);
4712 } else {
4713 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4714 }
4715 /* don't need this routine any more */
4716 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4717 }
4718 return;
4719 reschedule:
4720 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4721 mod_timer(&adapter->sfp_timer,
4722 round_jiffies(jiffies + (2 * HZ)));
4723 }
4724
4725 /**
4726 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4727 * @adapter: board private structure to initialize
4728 *
4729 * ixgbe_sw_init initializes the Adapter private data structure.
4730 * Fields are initialized based on PCI device information and
4731 * OS network device settings (MTU size).
4732 **/
4733 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4734 {
4735 struct ixgbe_hw *hw = &adapter->hw;
4736 struct pci_dev *pdev = adapter->pdev;
4737 struct net_device *dev = adapter->netdev;
4738 unsigned int rss;
4739 #ifdef CONFIG_IXGBE_DCB
4740 int j;
4741 struct tc_configuration *tc;
4742 #endif
4743
4744 /* PCI config space info */
4745
4746 hw->vendor_id = pdev->vendor;
4747 hw->device_id = pdev->device;
4748 hw->revision_id = pdev->revision;
4749 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4750 hw->subsystem_device_id = pdev->subsystem_device;
4751
4752 /* Set capability flags */
4753 rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4754 adapter->ring_feature[RING_F_RSS].indices = rss;
4755 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4756 adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4757 if (hw->mac.type == ixgbe_mac_82598EB) {
4758 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4759 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4760 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4761 } else if (hw->mac.type == ixgbe_mac_82599EB) {
4762 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4763 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4764 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4765 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4766 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4767 if (dev->features & NETIF_F_NTUPLE) {
4768 /* Flow Director perfect filter enabled */
4769 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4770 adapter->atr_sample_rate = 0;
4771 spin_lock_init(&adapter->fdir_perfect_lock);
4772 } else {
4773 /* Flow Director hash filters enabled */
4774 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4775 adapter->atr_sample_rate = 20;
4776 }
4777 adapter->ring_feature[RING_F_FDIR].indices =
4778 IXGBE_MAX_FDIR_INDICES;
4779 adapter->fdir_pballoc = 0;
4780 #ifdef IXGBE_FCOE
4781 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4782 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4783 adapter->ring_feature[RING_F_FCOE].indices = 0;
4784 #ifdef CONFIG_IXGBE_DCB
4785 /* Default traffic class to use for FCoE */
4786 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4787 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4788 #endif
4789 #endif /* IXGBE_FCOE */
4790 }
4791
4792 #ifdef CONFIG_IXGBE_DCB
4793 /* Configure DCB traffic classes */
4794 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4795 tc = &adapter->dcb_cfg.tc_config[j];
4796 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4797 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4798 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4799 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4800 tc->dcb_pfc = pfc_disabled;
4801 }
4802 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4803 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4804 adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4805 adapter->dcb_cfg.pfc_mode_enable = false;
4806 adapter->dcb_cfg.round_robin_enable = false;
4807 adapter->dcb_set_bitmap = 0x00;
4808 ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4809 adapter->ring_feature[RING_F_DCB].indices);
4810
4811 #endif
4812
4813 /* default flow control settings */
4814 hw->fc.requested_mode = ixgbe_fc_full;
4815 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4816 #ifdef CONFIG_DCB
4817 adapter->last_lfc_mode = hw->fc.current_mode;
4818 #endif
4819 hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4820 hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4821 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4822 hw->fc.send_xon = true;
4823 hw->fc.disable_fc_autoneg = false;
4824
4825 /* enable itr by default in dynamic mode */
4826 adapter->rx_itr_setting = 1;
4827 adapter->rx_eitr_param = 20000;
4828 adapter->tx_itr_setting = 1;
4829 adapter->tx_eitr_param = 10000;
4830
4831 /* set defaults for eitr in MegaBytes */
4832 adapter->eitr_low = 10;
4833 adapter->eitr_high = 20;
4834
4835 /* set default ring sizes */
4836 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4837 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4838
4839 /* initialize eeprom parameters */
4840 if (ixgbe_init_eeprom_params_generic(hw)) {
4841 e_dev_err("EEPROM initialization failed\n");
4842 return -EIO;
4843 }
4844
4845 /* enable rx csum by default */
4846 adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4847
4848 /* get assigned NUMA node */
4849 adapter->node = dev_to_node(&pdev->dev);
4850
4851 set_bit(__IXGBE_DOWN, &adapter->state);
4852
4853 return 0;
4854 }
4855
4856 /**
4857 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4858 * @adapter: board private structure
4859 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4860 *
4861 * Return 0 on success, negative on failure
4862 **/
4863 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4864 struct ixgbe_ring *tx_ring)
4865 {
4866 struct pci_dev *pdev = adapter->pdev;
4867 int size;
4868
4869 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4870 tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4871 if (!tx_ring->tx_buffer_info)
4872 tx_ring->tx_buffer_info = vmalloc(size);
4873 if (!tx_ring->tx_buffer_info)
4874 goto err;
4875 memset(tx_ring->tx_buffer_info, 0, size);
4876
4877 /* round up to nearest 4K */
4878 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4879 tx_ring->size = ALIGN(tx_ring->size, 4096);
4880
4881 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4882 &tx_ring->dma, GFP_KERNEL);
4883 if (!tx_ring->desc)
4884 goto err;
4885
4886 tx_ring->next_to_use = 0;
4887 tx_ring->next_to_clean = 0;
4888 tx_ring->work_limit = tx_ring->count;
4889 return 0;
4890
4891 err:
4892 vfree(tx_ring->tx_buffer_info);
4893 tx_ring->tx_buffer_info = NULL;
4894 e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
4895 return -ENOMEM;
4896 }
4897
4898 /**
4899 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4900 * @adapter: board private structure
4901 *
4902 * If this function returns with an error, then it's possible one or
4903 * more of the rings is populated (while the rest are not). It is the
4904 * callers duty to clean those orphaned rings.
4905 *
4906 * Return 0 on success, negative on failure
4907 **/
4908 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4909 {
4910 int i, err = 0;
4911
4912 for (i = 0; i < adapter->num_tx_queues; i++) {
4913 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4914 if (!err)
4915 continue;
4916 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4917 break;
4918 }
4919
4920 return err;
4921 }
4922
4923 /**
4924 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4925 * @adapter: board private structure
4926 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4927 *
4928 * Returns 0 on success, negative on failure
4929 **/
4930 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4931 struct ixgbe_ring *rx_ring)
4932 {
4933 struct pci_dev *pdev = adapter->pdev;
4934 int size;
4935
4936 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4937 rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
4938 if (!rx_ring->rx_buffer_info)
4939 rx_ring->rx_buffer_info = vmalloc(size);
4940 if (!rx_ring->rx_buffer_info) {
4941 e_err(probe, "vmalloc allocation failed for the Rx "
4942 "descriptor ring\n");
4943 goto alloc_failed;
4944 }
4945 memset(rx_ring->rx_buffer_info, 0, size);
4946
4947 /* Round up to nearest 4K */
4948 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4949 rx_ring->size = ALIGN(rx_ring->size, 4096);
4950
4951 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
4952 &rx_ring->dma, GFP_KERNEL);
4953
4954 if (!rx_ring->desc) {
4955 e_err(probe, "Memory allocation failed for the Rx "
4956 "descriptor ring\n");
4957 vfree(rx_ring->rx_buffer_info);
4958 goto alloc_failed;
4959 }
4960
4961 rx_ring->next_to_clean = 0;
4962 rx_ring->next_to_use = 0;
4963
4964 return 0;
4965
4966 alloc_failed:
4967 return -ENOMEM;
4968 }
4969
4970 /**
4971 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4972 * @adapter: board private structure
4973 *
4974 * If this function returns with an error, then it's possible one or
4975 * more of the rings is populated (while the rest are not). It is the
4976 * callers duty to clean those orphaned rings.
4977 *
4978 * Return 0 on success, negative on failure
4979 **/
4980
4981 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4982 {
4983 int i, err = 0;
4984
4985 for (i = 0; i < adapter->num_rx_queues; i++) {
4986 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
4987 if (!err)
4988 continue;
4989 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4990 break;
4991 }
4992
4993 return err;
4994 }
4995
4996 /**
4997 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4998 * @adapter: board private structure
4999 * @tx_ring: Tx descriptor ring for a specific queue
5000 *
5001 * Free all transmit software resources
5002 **/
5003 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
5004 struct ixgbe_ring *tx_ring)
5005 {
5006 struct pci_dev *pdev = adapter->pdev;
5007
5008 ixgbe_clean_tx_ring(adapter, tx_ring);
5009
5010 vfree(tx_ring->tx_buffer_info);
5011 tx_ring->tx_buffer_info = NULL;
5012
5013 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5014 tx_ring->dma);
5015
5016 tx_ring->desc = NULL;
5017 }
5018
5019 /**
5020 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5021 * @adapter: board private structure
5022 *
5023 * Free all transmit software resources
5024 **/
5025 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5026 {
5027 int i;
5028
5029 for (i = 0; i < adapter->num_tx_queues; i++)
5030 if (adapter->tx_ring[i]->desc)
5031 ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5032 }
5033
5034 /**
5035 * ixgbe_free_rx_resources - Free Rx Resources
5036 * @adapter: board private structure
5037 * @rx_ring: ring to clean the resources from
5038 *
5039 * Free all receive software resources
5040 **/
5041 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5042 struct ixgbe_ring *rx_ring)
5043 {
5044 struct pci_dev *pdev = adapter->pdev;
5045
5046 ixgbe_clean_rx_ring(adapter, rx_ring);
5047
5048 vfree(rx_ring->rx_buffer_info);
5049 rx_ring->rx_buffer_info = NULL;
5050
5051 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5052 rx_ring->dma);
5053
5054 rx_ring->desc = NULL;
5055 }
5056
5057 /**
5058 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5059 * @adapter: board private structure
5060 *
5061 * Free all receive software resources
5062 **/
5063 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5064 {
5065 int i;
5066
5067 for (i = 0; i < adapter->num_rx_queues; i++)
5068 if (adapter->rx_ring[i]->desc)
5069 ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5070 }
5071
5072 /**
5073 * ixgbe_change_mtu - Change the Maximum Transfer Unit
5074 * @netdev: network interface device structure
5075 * @new_mtu: new value for maximum frame size
5076 *
5077 * Returns 0 on success, negative on failure
5078 **/
5079 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5080 {
5081 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5082 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5083
5084 /* MTU < 68 is an error and causes problems on some kernels */
5085 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5086 return -EINVAL;
5087
5088 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5089 /* must set new MTU before calling down or up */
5090 netdev->mtu = new_mtu;
5091
5092 if (netif_running(netdev))
5093 ixgbe_reinit_locked(adapter);
5094
5095 return 0;
5096 }
5097
5098 /**
5099 * ixgbe_open - Called when a network interface is made active
5100 * @netdev: network interface device structure
5101 *
5102 * Returns 0 on success, negative value on failure
5103 *
5104 * The open entry point is called when a network interface is made
5105 * active by the system (IFF_UP). At this point all resources needed
5106 * for transmit and receive operations are allocated, the interrupt
5107 * handler is registered with the OS, the watchdog timer is started,
5108 * and the stack is notified that the interface is ready.
5109 **/
5110 static int ixgbe_open(struct net_device *netdev)
5111 {
5112 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5113 int err;
5114
5115 /* disallow open during test */
5116 if (test_bit(__IXGBE_TESTING, &adapter->state))
5117 return -EBUSY;
5118
5119 netif_carrier_off(netdev);
5120
5121 /* allocate transmit descriptors */
5122 err = ixgbe_setup_all_tx_resources(adapter);
5123 if (err)
5124 goto err_setup_tx;
5125
5126 /* allocate receive descriptors */
5127 err = ixgbe_setup_all_rx_resources(adapter);
5128 if (err)
5129 goto err_setup_rx;
5130
5131 ixgbe_configure(adapter);
5132
5133 err = ixgbe_request_irq(adapter);
5134 if (err)
5135 goto err_req_irq;
5136
5137 err = ixgbe_up_complete(adapter);
5138 if (err)
5139 goto err_up;
5140
5141 netif_tx_start_all_queues(netdev);
5142
5143 return 0;
5144
5145 err_up:
5146 ixgbe_release_hw_control(adapter);
5147 ixgbe_free_irq(adapter);
5148 err_req_irq:
5149 err_setup_rx:
5150 ixgbe_free_all_rx_resources(adapter);
5151 err_setup_tx:
5152 ixgbe_free_all_tx_resources(adapter);
5153 ixgbe_reset(adapter);
5154
5155 return err;
5156 }
5157
5158 /**
5159 * ixgbe_close - Disables a network interface
5160 * @netdev: network interface device structure
5161 *
5162 * Returns 0, this is not allowed to fail
5163 *
5164 * The close entry point is called when an interface is de-activated
5165 * by the OS. The hardware is still under the drivers control, but
5166 * needs to be disabled. A global MAC reset is issued to stop the
5167 * hardware, and all transmit and receive resources are freed.
5168 **/
5169 static int ixgbe_close(struct net_device *netdev)
5170 {
5171 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5172
5173 ixgbe_down(adapter);
5174 ixgbe_free_irq(adapter);
5175
5176 ixgbe_free_all_tx_resources(adapter);
5177 ixgbe_free_all_rx_resources(adapter);
5178
5179 ixgbe_release_hw_control(adapter);
5180
5181 return 0;
5182 }
5183
5184 #ifdef CONFIG_PM
5185 static int ixgbe_resume(struct pci_dev *pdev)
5186 {
5187 struct net_device *netdev = pci_get_drvdata(pdev);
5188 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5189 u32 err;
5190
5191 pci_set_power_state(pdev, PCI_D0);
5192 pci_restore_state(pdev);
5193 /*
5194 * pci_restore_state clears dev->state_saved so call
5195 * pci_save_state to restore it.
5196 */
5197 pci_save_state(pdev);
5198
5199 err = pci_enable_device_mem(pdev);
5200 if (err) {
5201 e_dev_err("Cannot enable PCI device from suspend\n");
5202 return err;
5203 }
5204 pci_set_master(pdev);
5205
5206 pci_wake_from_d3(pdev, false);
5207
5208 err = ixgbe_init_interrupt_scheme(adapter);
5209 if (err) {
5210 e_dev_err("Cannot initialize interrupts for device\n");
5211 return err;
5212 }
5213
5214 ixgbe_reset(adapter);
5215
5216 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5217
5218 if (netif_running(netdev)) {
5219 err = ixgbe_open(adapter->netdev);
5220 if (err)
5221 return err;
5222 }
5223
5224 netif_device_attach(netdev);
5225
5226 return 0;
5227 }
5228 #endif /* CONFIG_PM */
5229
5230 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5231 {
5232 struct net_device *netdev = pci_get_drvdata(pdev);
5233 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5234 struct ixgbe_hw *hw = &adapter->hw;
5235 u32 ctrl, fctrl;
5236 u32 wufc = adapter->wol;
5237 #ifdef CONFIG_PM
5238 int retval = 0;
5239 #endif
5240
5241 netif_device_detach(netdev);
5242
5243 if (netif_running(netdev)) {
5244 ixgbe_down(adapter);
5245 ixgbe_free_irq(adapter);
5246 ixgbe_free_all_tx_resources(adapter);
5247 ixgbe_free_all_rx_resources(adapter);
5248 }
5249
5250 #ifdef CONFIG_PM
5251 retval = pci_save_state(pdev);
5252 if (retval)
5253 return retval;
5254
5255 #endif
5256 if (wufc) {
5257 ixgbe_set_rx_mode(netdev);
5258
5259 /* turn on all-multi mode if wake on multicast is enabled */
5260 if (wufc & IXGBE_WUFC_MC) {
5261 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5262 fctrl |= IXGBE_FCTRL_MPE;
5263 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5264 }
5265
5266 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5267 ctrl |= IXGBE_CTRL_GIO_DIS;
5268 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5269
5270 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5271 } else {
5272 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5273 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5274 }
5275
5276 if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5277 pci_wake_from_d3(pdev, true);
5278 else
5279 pci_wake_from_d3(pdev, false);
5280
5281 *enable_wake = !!wufc;
5282
5283 ixgbe_clear_interrupt_scheme(adapter);
5284
5285 ixgbe_release_hw_control(adapter);
5286
5287 pci_disable_device(pdev);
5288
5289 return 0;
5290 }
5291
5292 #ifdef CONFIG_PM
5293 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5294 {
5295 int retval;
5296 bool wake;
5297
5298 retval = __ixgbe_shutdown(pdev, &wake);
5299 if (retval)
5300 return retval;
5301
5302 if (wake) {
5303 pci_prepare_to_sleep(pdev);
5304 } else {
5305 pci_wake_from_d3(pdev, false);
5306 pci_set_power_state(pdev, PCI_D3hot);
5307 }
5308
5309 return 0;
5310 }
5311 #endif /* CONFIG_PM */
5312
5313 static void ixgbe_shutdown(struct pci_dev *pdev)
5314 {
5315 bool wake;
5316
5317 __ixgbe_shutdown(pdev, &wake);
5318
5319 if (system_state == SYSTEM_POWER_OFF) {
5320 pci_wake_from_d3(pdev, wake);
5321 pci_set_power_state(pdev, PCI_D3hot);
5322 }
5323 }
5324
5325 /**
5326 * ixgbe_update_stats - Update the board statistics counters.
5327 * @adapter: board private structure
5328 **/
5329 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5330 {
5331 struct net_device *netdev = adapter->netdev;
5332 struct ixgbe_hw *hw = &adapter->hw;
5333 u64 total_mpc = 0;
5334 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5335 u64 non_eop_descs = 0, restart_queue = 0;
5336
5337 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5338 test_bit(__IXGBE_RESETTING, &adapter->state))
5339 return;
5340
5341 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5342 u64 rsc_count = 0;
5343 u64 rsc_flush = 0;
5344 for (i = 0; i < 16; i++)
5345 adapter->hw_rx_no_dma_resources +=
5346 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5347 for (i = 0; i < adapter->num_rx_queues; i++) {
5348 rsc_count += adapter->rx_ring[i]->rsc_count;
5349 rsc_flush += adapter->rx_ring[i]->rsc_flush;
5350 }
5351 adapter->rsc_total_count = rsc_count;
5352 adapter->rsc_total_flush = rsc_flush;
5353 }
5354
5355 /* gather some stats to the adapter struct that are per queue */
5356 for (i = 0; i < adapter->num_tx_queues; i++)
5357 restart_queue += adapter->tx_ring[i]->restart_queue;
5358 adapter->restart_queue = restart_queue;
5359
5360 for (i = 0; i < adapter->num_rx_queues; i++)
5361 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5362 adapter->non_eop_descs = non_eop_descs;
5363
5364 adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5365 for (i = 0; i < 8; i++) {
5366 /* for packet buffers not used, the register should read 0 */
5367 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5368 missed_rx += mpc;
5369 adapter->stats.mpc[i] += mpc;
5370 total_mpc += adapter->stats.mpc[i];
5371 if (hw->mac.type == ixgbe_mac_82598EB)
5372 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5373 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5374 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5375 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5376 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5377 if (hw->mac.type == ixgbe_mac_82599EB) {
5378 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5379 IXGBE_PXONRXCNT(i));
5380 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5381 IXGBE_PXOFFRXCNT(i));
5382 adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5383 } else {
5384 adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5385 IXGBE_PXONRXC(i));
5386 adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5387 IXGBE_PXOFFRXC(i));
5388 }
5389 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5390 IXGBE_PXONTXC(i));
5391 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5392 IXGBE_PXOFFTXC(i));
5393 }
5394 adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5395 /* work around hardware counting issue */
5396 adapter->stats.gprc -= missed_rx;
5397
5398 /* 82598 hardware only has a 32 bit counter in the high register */
5399 if (hw->mac.type == ixgbe_mac_82599EB) {
5400 u64 tmp;
5401 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5402 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
5403 adapter->stats.gorc += (tmp << 32);
5404 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5405 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
5406 adapter->stats.gotc += (tmp << 32);
5407 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5408 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5409 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5410 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5411 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5412 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5413 #ifdef IXGBE_FCOE
5414 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5415 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5416 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5417 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5418 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5419 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5420 #endif /* IXGBE_FCOE */
5421 } else {
5422 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5423 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5424 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5425 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5426 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5427 }
5428 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5429 adapter->stats.bprc += bprc;
5430 adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5431 if (hw->mac.type == ixgbe_mac_82598EB)
5432 adapter->stats.mprc -= bprc;
5433 adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5434 adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5435 adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5436 adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5437 adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5438 adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5439 adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5440 adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5441 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5442 adapter->stats.lxontxc += lxon;
5443 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5444 adapter->stats.lxofftxc += lxoff;
5445 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5446 adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5447 adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5448 /*
5449 * 82598 errata - tx of flow control packets is included in tx counters
5450 */
5451 xon_off_tot = lxon + lxoff;
5452 adapter->stats.gptc -= xon_off_tot;
5453 adapter->stats.mptc -= xon_off_tot;
5454 adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5455 adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5456 adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5457 adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5458 adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5459 adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5460 adapter->stats.ptc64 -= xon_off_tot;
5461 adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5462 adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5463 adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5464 adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5465 adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5466 adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5467
5468 /* Fill out the OS statistics structure */
5469 netdev->stats.multicast = adapter->stats.mprc;
5470
5471 /* Rx Errors */
5472 netdev->stats.rx_errors = adapter->stats.crcerrs +
5473 adapter->stats.rlec;
5474 netdev->stats.rx_dropped = 0;
5475 netdev->stats.rx_length_errors = adapter->stats.rlec;
5476 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5477 netdev->stats.rx_missed_errors = total_mpc;
5478 }
5479
5480 /**
5481 * ixgbe_watchdog - Timer Call-back
5482 * @data: pointer to adapter cast into an unsigned long
5483 **/
5484 static void ixgbe_watchdog(unsigned long data)
5485 {
5486 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5487 struct ixgbe_hw *hw = &adapter->hw;
5488 u64 eics = 0;
5489 int i;
5490
5491 /*
5492 * Do the watchdog outside of interrupt context due to the lovely
5493 * delays that some of the newer hardware requires
5494 */
5495
5496 if (test_bit(__IXGBE_DOWN, &adapter->state))
5497 goto watchdog_short_circuit;
5498
5499 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5500 /*
5501 * for legacy and MSI interrupts don't set any bits
5502 * that are enabled for EIAM, because this operation
5503 * would set *both* EIMS and EICS for any bit in EIAM
5504 */
5505 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5506 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5507 goto watchdog_reschedule;
5508 }
5509
5510 /* get one bit for every active tx/rx interrupt vector */
5511 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5512 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5513 if (qv->rxr_count || qv->txr_count)
5514 eics |= ((u64)1 << i);
5515 }
5516
5517 /* Cause software interrupt to ensure rx rings are cleaned */
5518 ixgbe_irq_rearm_queues(adapter, eics);
5519
5520 watchdog_reschedule:
5521 /* Reset the timer */
5522 mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5523
5524 watchdog_short_circuit:
5525 schedule_work(&adapter->watchdog_task);
5526 }
5527
5528 /**
5529 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5530 * @work: pointer to work_struct containing our data
5531 **/
5532 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5533 {
5534 struct ixgbe_adapter *adapter = container_of(work,
5535 struct ixgbe_adapter,
5536 multispeed_fiber_task);
5537 struct ixgbe_hw *hw = &adapter->hw;
5538 u32 autoneg;
5539 bool negotiation;
5540
5541 adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5542 autoneg = hw->phy.autoneg_advertised;
5543 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5544 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5545 hw->mac.autotry_restart = false;
5546 if (hw->mac.ops.setup_link)
5547 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5548 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5549 adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5550 }
5551
5552 /**
5553 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5554 * @work: pointer to work_struct containing our data
5555 **/
5556 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5557 {
5558 struct ixgbe_adapter *adapter = container_of(work,
5559 struct ixgbe_adapter,
5560 sfp_config_module_task);
5561 struct ixgbe_hw *hw = &adapter->hw;
5562 u32 err;
5563
5564 adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5565
5566 /* Time for electrical oscillations to settle down */
5567 msleep(100);
5568 err = hw->phy.ops.identify_sfp(hw);
5569
5570 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5571 e_dev_err("failed to initialize because an unsupported SFP+ "
5572 "module type was detected.\n");
5573 e_dev_err("Reload the driver after installing a supported "
5574 "module.\n");
5575 unregister_netdev(adapter->netdev);
5576 return;
5577 }
5578 hw->mac.ops.setup_sfp(hw);
5579
5580 if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5581 /* This will also work for DA Twinax connections */
5582 schedule_work(&adapter->multispeed_fiber_task);
5583 adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5584 }
5585
5586 /**
5587 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5588 * @work: pointer to work_struct containing our data
5589 **/
5590 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5591 {
5592 struct ixgbe_adapter *adapter = container_of(work,
5593 struct ixgbe_adapter,
5594 fdir_reinit_task);
5595 struct ixgbe_hw *hw = &adapter->hw;
5596 int i;
5597
5598 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5599 for (i = 0; i < adapter->num_tx_queues; i++)
5600 set_bit(__IXGBE_FDIR_INIT_DONE,
5601 &(adapter->tx_ring[i]->reinit_state));
5602 } else {
5603 e_err(probe, "failed to finish FDIR re-initialization, "
5604 "ignored adding FDIR ATR filters\n");
5605 }
5606 /* Done FDIR Re-initialization, enable transmits */
5607 netif_tx_start_all_queues(adapter->netdev);
5608 }
5609
5610 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5611
5612 /**
5613 * ixgbe_watchdog_task - worker thread to bring link up
5614 * @work: pointer to work_struct containing our data
5615 **/
5616 static void ixgbe_watchdog_task(struct work_struct *work)
5617 {
5618 struct ixgbe_adapter *adapter = container_of(work,
5619 struct ixgbe_adapter,
5620 watchdog_task);
5621 struct net_device *netdev = adapter->netdev;
5622 struct ixgbe_hw *hw = &adapter->hw;
5623 u32 link_speed;
5624 bool link_up;
5625 int i;
5626 struct ixgbe_ring *tx_ring;
5627 int some_tx_pending = 0;
5628
5629 mutex_lock(&ixgbe_watchdog_lock);
5630
5631 link_up = adapter->link_up;
5632 link_speed = adapter->link_speed;
5633
5634 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5635 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5636 if (link_up) {
5637 #ifdef CONFIG_DCB
5638 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5639 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5640 hw->mac.ops.fc_enable(hw, i);
5641 } else {
5642 hw->mac.ops.fc_enable(hw, 0);
5643 }
5644 #else
5645 hw->mac.ops.fc_enable(hw, 0);
5646 #endif
5647 }
5648
5649 if (link_up ||
5650 time_after(jiffies, (adapter->link_check_timeout +
5651 IXGBE_TRY_LINK_TIMEOUT))) {
5652 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5653 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5654 }
5655 adapter->link_up = link_up;
5656 adapter->link_speed = link_speed;
5657 }
5658
5659 if (link_up) {
5660 if (!netif_carrier_ok(netdev)) {
5661 bool flow_rx, flow_tx;
5662
5663 if (hw->mac.type == ixgbe_mac_82599EB) {
5664 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5665 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5666 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5667 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5668 } else {
5669 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5670 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5671 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5672 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5673 }
5674
5675 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5676 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5677 "10 Gbps" :
5678 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5679 "1 Gbps" : "unknown speed")),
5680 ((flow_rx && flow_tx) ? "RX/TX" :
5681 (flow_rx ? "RX" :
5682 (flow_tx ? "TX" : "None"))));
5683
5684 netif_carrier_on(netdev);
5685 } else {
5686 /* Force detection of hung controller */
5687 adapter->detect_tx_hung = true;
5688 }
5689 } else {
5690 adapter->link_up = false;
5691 adapter->link_speed = 0;
5692 if (netif_carrier_ok(netdev)) {
5693 e_info(drv, "NIC Link is Down\n");
5694 netif_carrier_off(netdev);
5695 }
5696 }
5697
5698 if (!netif_carrier_ok(netdev)) {
5699 for (i = 0; i < adapter->num_tx_queues; i++) {
5700 tx_ring = adapter->tx_ring[i];
5701 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5702 some_tx_pending = 1;
5703 break;
5704 }
5705 }
5706
5707 if (some_tx_pending) {
5708 /* We've lost link, so the controller stops DMA,
5709 * but we've got queued Tx work that's never going
5710 * to get done, so reset controller to flush Tx.
5711 * (Do the reset outside of interrupt context).
5712 */
5713 schedule_work(&adapter->reset_task);
5714 }
5715 }
5716
5717 ixgbe_update_stats(adapter);
5718 mutex_unlock(&ixgbe_watchdog_lock);
5719 }
5720
5721 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5722 struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5723 u32 tx_flags, u8 *hdr_len)
5724 {
5725 struct ixgbe_adv_tx_context_desc *context_desc;
5726 unsigned int i;
5727 int err;
5728 struct ixgbe_tx_buffer *tx_buffer_info;
5729 u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5730 u32 mss_l4len_idx, l4len;
5731
5732 if (skb_is_gso(skb)) {
5733 if (skb_header_cloned(skb)) {
5734 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5735 if (err)
5736 return err;
5737 }
5738 l4len = tcp_hdrlen(skb);
5739 *hdr_len += l4len;
5740
5741 if (skb->protocol == htons(ETH_P_IP)) {
5742 struct iphdr *iph = ip_hdr(skb);
5743 iph->tot_len = 0;
5744 iph->check = 0;
5745 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5746 iph->daddr, 0,
5747 IPPROTO_TCP,
5748 0);
5749 } else if (skb_is_gso_v6(skb)) {
5750 ipv6_hdr(skb)->payload_len = 0;
5751 tcp_hdr(skb)->check =
5752 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5753 &ipv6_hdr(skb)->daddr,
5754 0, IPPROTO_TCP, 0);
5755 }
5756
5757 i = tx_ring->next_to_use;
5758
5759 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5760 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5761
5762 /* VLAN MACLEN IPLEN */
5763 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5764 vlan_macip_lens |=
5765 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5766 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5767 IXGBE_ADVTXD_MACLEN_SHIFT);
5768 *hdr_len += skb_network_offset(skb);
5769 vlan_macip_lens |=
5770 (skb_transport_header(skb) - skb_network_header(skb));
5771 *hdr_len +=
5772 (skb_transport_header(skb) - skb_network_header(skb));
5773 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5774 context_desc->seqnum_seed = 0;
5775
5776 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5777 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5778 IXGBE_ADVTXD_DTYP_CTXT);
5779
5780 if (skb->protocol == htons(ETH_P_IP))
5781 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5782 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5783 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5784
5785 /* MSS L4LEN IDX */
5786 mss_l4len_idx =
5787 (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5788 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5789 /* use index 1 for TSO */
5790 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5791 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5792
5793 tx_buffer_info->time_stamp = jiffies;
5794 tx_buffer_info->next_to_watch = i;
5795
5796 i++;
5797 if (i == tx_ring->count)
5798 i = 0;
5799 tx_ring->next_to_use = i;
5800
5801 return true;
5802 }
5803 return false;
5804 }
5805
5806 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5807 struct ixgbe_ring *tx_ring,
5808 struct sk_buff *skb, u32 tx_flags)
5809 {
5810 struct ixgbe_adv_tx_context_desc *context_desc;
5811 unsigned int i;
5812 struct ixgbe_tx_buffer *tx_buffer_info;
5813 u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5814
5815 if (skb->ip_summed == CHECKSUM_PARTIAL ||
5816 (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5817 i = tx_ring->next_to_use;
5818 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5819 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
5820
5821 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5822 vlan_macip_lens |=
5823 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5824 vlan_macip_lens |= (skb_network_offset(skb) <<
5825 IXGBE_ADVTXD_MACLEN_SHIFT);
5826 if (skb->ip_summed == CHECKSUM_PARTIAL)
5827 vlan_macip_lens |= (skb_transport_header(skb) -
5828 skb_network_header(skb));
5829
5830 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5831 context_desc->seqnum_seed = 0;
5832
5833 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5834 IXGBE_ADVTXD_DTYP_CTXT);
5835
5836 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5837 __be16 protocol;
5838
5839 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5840 const struct vlan_ethhdr *vhdr =
5841 (const struct vlan_ethhdr *)skb->data;
5842
5843 protocol = vhdr->h_vlan_encapsulated_proto;
5844 } else {
5845 protocol = skb->protocol;
5846 }
5847
5848 switch (protocol) {
5849 case cpu_to_be16(ETH_P_IP):
5850 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5851 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5852 type_tucmd_mlhl |=
5853 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5854 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5855 type_tucmd_mlhl |=
5856 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5857 break;
5858 case cpu_to_be16(ETH_P_IPV6):
5859 /* XXX what about other V6 headers?? */
5860 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5861 type_tucmd_mlhl |=
5862 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5863 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5864 type_tucmd_mlhl |=
5865 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5866 break;
5867 default:
5868 if (unlikely(net_ratelimit())) {
5869 e_warn(probe, "partial checksum "
5870 "but proto=%x!\n",
5871 skb->protocol);
5872 }
5873 break;
5874 }
5875 }
5876
5877 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5878 /* use index zero for tx checksum offload */
5879 context_desc->mss_l4len_idx = 0;
5880
5881 tx_buffer_info->time_stamp = jiffies;
5882 tx_buffer_info->next_to_watch = i;
5883
5884 i++;
5885 if (i == tx_ring->count)
5886 i = 0;
5887 tx_ring->next_to_use = i;
5888
5889 return true;
5890 }
5891
5892 return false;
5893 }
5894
5895 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5896 struct ixgbe_ring *tx_ring,
5897 struct sk_buff *skb, u32 tx_flags,
5898 unsigned int first)
5899 {
5900 struct pci_dev *pdev = adapter->pdev;
5901 struct ixgbe_tx_buffer *tx_buffer_info;
5902 unsigned int len;
5903 unsigned int total = skb->len;
5904 unsigned int offset = 0, size, count = 0, i;
5905 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5906 unsigned int f;
5907
5908 i = tx_ring->next_to_use;
5909
5910 if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5911 /* excluding fcoe_crc_eof for FCoE */
5912 total -= sizeof(struct fcoe_crc_eof);
5913
5914 len = min(skb_headlen(skb), total);
5915 while (len) {
5916 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5917 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5918
5919 tx_buffer_info->length = size;
5920 tx_buffer_info->mapped_as_page = false;
5921 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5922 skb->data + offset,
5923 size, DMA_TO_DEVICE);
5924 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5925 goto dma_error;
5926 tx_buffer_info->time_stamp = jiffies;
5927 tx_buffer_info->next_to_watch = i;
5928
5929 len -= size;
5930 total -= size;
5931 offset += size;
5932 count++;
5933
5934 if (len) {
5935 i++;
5936 if (i == tx_ring->count)
5937 i = 0;
5938 }
5939 }
5940
5941 for (f = 0; f < nr_frags; f++) {
5942 struct skb_frag_struct *frag;
5943
5944 frag = &skb_shinfo(skb)->frags[f];
5945 len = min((unsigned int)frag->size, total);
5946 offset = frag->page_offset;
5947
5948 while (len) {
5949 i++;
5950 if (i == tx_ring->count)
5951 i = 0;
5952
5953 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5954 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5955
5956 tx_buffer_info->length = size;
5957 tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
5958 frag->page,
5959 offset, size,
5960 DMA_TO_DEVICE);
5961 tx_buffer_info->mapped_as_page = true;
5962 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5963 goto dma_error;
5964 tx_buffer_info->time_stamp = jiffies;
5965 tx_buffer_info->next_to_watch = i;
5966
5967 len -= size;
5968 total -= size;
5969 offset += size;
5970 count++;
5971 }
5972 if (total == 0)
5973 break;
5974 }
5975
5976 tx_ring->tx_buffer_info[i].skb = skb;
5977 tx_ring->tx_buffer_info[first].next_to_watch = i;
5978
5979 return count;
5980
5981 dma_error:
5982 e_dev_err("TX DMA map failed\n");
5983
5984 /* clear timestamp and dma mappings for failed tx_buffer_info map */
5985 tx_buffer_info->dma = 0;
5986 tx_buffer_info->time_stamp = 0;
5987 tx_buffer_info->next_to_watch = 0;
5988 if (count)
5989 count--;
5990
5991 /* clear timestamp and dma mappings for remaining portion of packet */
5992 while (count--) {
5993 if (i==0)
5994 i += tx_ring->count;
5995 i--;
5996 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5997 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
5998 }
5999
6000 return 0;
6001 }
6002
6003 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
6004 struct ixgbe_ring *tx_ring,
6005 int tx_flags, int count, u32 paylen, u8 hdr_len)
6006 {
6007 union ixgbe_adv_tx_desc *tx_desc = NULL;
6008 struct ixgbe_tx_buffer *tx_buffer_info;
6009 u32 olinfo_status = 0, cmd_type_len = 0;
6010 unsigned int i;
6011 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6012
6013 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6014
6015 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6016
6017 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6018 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6019
6020 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6021 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6022
6023 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6024 IXGBE_ADVTXD_POPTS_SHIFT;
6025
6026 /* use index 1 context for tso */
6027 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6028 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6029 olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6030 IXGBE_ADVTXD_POPTS_SHIFT;
6031
6032 } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6033 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6034 IXGBE_ADVTXD_POPTS_SHIFT;
6035
6036 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6037 olinfo_status |= IXGBE_ADVTXD_CC;
6038 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6039 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6040 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6041 }
6042
6043 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6044
6045 i = tx_ring->next_to_use;
6046 while (count--) {
6047 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6048 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
6049 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6050 tx_desc->read.cmd_type_len =
6051 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6052 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6053 i++;
6054 if (i == tx_ring->count)
6055 i = 0;
6056 }
6057
6058 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6059
6060 /*
6061 * Force memory writes to complete before letting h/w
6062 * know there are new descriptors to fetch. (Only
6063 * applicable for weak-ordered memory model archs,
6064 * such as IA-64).
6065 */
6066 wmb();
6067
6068 tx_ring->next_to_use = i;
6069 writel(i, adapter->hw.hw_addr + tx_ring->tail);
6070 }
6071
6072 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6073 int queue, u32 tx_flags)
6074 {
6075 struct ixgbe_atr_input atr_input;
6076 struct tcphdr *th;
6077 struct iphdr *iph = ip_hdr(skb);
6078 struct ethhdr *eth = (struct ethhdr *)skb->data;
6079 u16 vlan_id, src_port, dst_port, flex_bytes;
6080 u32 src_ipv4_addr, dst_ipv4_addr;
6081 u8 l4type = 0;
6082
6083 /* Right now, we support IPv4 only */
6084 if (skb->protocol != htons(ETH_P_IP))
6085 return;
6086 /* check if we're UDP or TCP */
6087 if (iph->protocol == IPPROTO_TCP) {
6088 th = tcp_hdr(skb);
6089 src_port = th->source;
6090 dst_port = th->dest;
6091 l4type |= IXGBE_ATR_L4TYPE_TCP;
6092 /* l4type IPv4 type is 0, no need to assign */
6093 } else {
6094 /* Unsupported L4 header, just bail here */
6095 return;
6096 }
6097
6098 memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6099
6100 vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6101 IXGBE_TX_FLAGS_VLAN_SHIFT;
6102 src_ipv4_addr = iph->saddr;
6103 dst_ipv4_addr = iph->daddr;
6104 flex_bytes = eth->h_proto;
6105
6106 ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6107 ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6108 ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6109 ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6110 ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6111 /* src and dst are inverted, think how the receiver sees them */
6112 ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6113 ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6114
6115 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6116 ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6117 }
6118
6119 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6120 struct ixgbe_ring *tx_ring, int size)
6121 {
6122 netif_stop_subqueue(netdev, tx_ring->queue_index);
6123 /* Herbert's original patch had:
6124 * smp_mb__after_netif_stop_queue();
6125 * but since that doesn't exist yet, just open code it. */
6126 smp_mb();
6127
6128 /* We need to check again in a case another CPU has just
6129 * made room available. */
6130 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6131 return -EBUSY;
6132
6133 /* A reprieve! - use start_queue because it doesn't call schedule */
6134 netif_start_subqueue(netdev, tx_ring->queue_index);
6135 ++tx_ring->restart_queue;
6136 return 0;
6137 }
6138
6139 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6140 struct ixgbe_ring *tx_ring, int size)
6141 {
6142 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6143 return 0;
6144 return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6145 }
6146
6147 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6148 {
6149 struct ixgbe_adapter *adapter = netdev_priv(dev);
6150 int txq = smp_processor_id();
6151
6152 #ifdef IXGBE_FCOE
6153 if ((skb->protocol == htons(ETH_P_FCOE)) ||
6154 (skb->protocol == htons(ETH_P_FIP))) {
6155 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6156 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6157 txq += adapter->ring_feature[RING_F_FCOE].mask;
6158 return txq;
6159 #ifdef CONFIG_IXGBE_DCB
6160 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6161 txq = adapter->fcoe.up;
6162 return txq;
6163 #endif
6164 }
6165 }
6166 #endif
6167
6168 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6169 while (unlikely(txq >= dev->real_num_tx_queues))
6170 txq -= dev->real_num_tx_queues;
6171 return txq;
6172 }
6173
6174 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6175 if (skb->priority == TC_PRIO_CONTROL)
6176 txq = adapter->ring_feature[RING_F_DCB].indices-1;
6177 else
6178 txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6179 >> 13;
6180 return txq;
6181 }
6182
6183 return skb_tx_hash(dev, skb);
6184 }
6185
6186 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6187 struct net_device *netdev)
6188 {
6189 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6190 struct ixgbe_ring *tx_ring;
6191 struct netdev_queue *txq;
6192 unsigned int first;
6193 unsigned int tx_flags = 0;
6194 u8 hdr_len = 0;
6195 int tso;
6196 int count = 0;
6197 unsigned int f;
6198
6199 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6200 tx_flags |= vlan_tx_tag_get(skb);
6201 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6202 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6203 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6204 }
6205 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6206 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6207 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6208 skb->priority != TC_PRIO_CONTROL) {
6209 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6210 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6211 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6212 }
6213
6214 tx_ring = adapter->tx_ring[skb->queue_mapping];
6215
6216 #ifdef IXGBE_FCOE
6217 /* for FCoE with DCB, we force the priority to what
6218 * was specified by the switch */
6219 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6220 (skb->protocol == htons(ETH_P_FCOE) ||
6221 skb->protocol == htons(ETH_P_FIP))) {
6222 #ifdef CONFIG_IXGBE_DCB
6223 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6224 tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6225 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6226 tx_flags |= ((adapter->fcoe.up << 13)
6227 << IXGBE_TX_FLAGS_VLAN_SHIFT);
6228 }
6229 #endif
6230 /* flag for FCoE offloads */
6231 if (skb->protocol == htons(ETH_P_FCOE))
6232 tx_flags |= IXGBE_TX_FLAGS_FCOE;
6233 }
6234 #endif
6235
6236 /* four things can cause us to need a context descriptor */
6237 if (skb_is_gso(skb) ||
6238 (skb->ip_summed == CHECKSUM_PARTIAL) ||
6239 (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6240 (tx_flags & IXGBE_TX_FLAGS_FCOE))
6241 count++;
6242
6243 count += TXD_USE_COUNT(skb_headlen(skb));
6244 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6245 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6246
6247 if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6248 adapter->tx_busy++;
6249 return NETDEV_TX_BUSY;
6250 }
6251
6252 first = tx_ring->next_to_use;
6253 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6254 #ifdef IXGBE_FCOE
6255 /* setup tx offload for FCoE */
6256 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6257 if (tso < 0) {
6258 dev_kfree_skb_any(skb);
6259 return NETDEV_TX_OK;
6260 }
6261 if (tso)
6262 tx_flags |= IXGBE_TX_FLAGS_FSO;
6263 #endif /* IXGBE_FCOE */
6264 } else {
6265 if (skb->protocol == htons(ETH_P_IP))
6266 tx_flags |= IXGBE_TX_FLAGS_IPV4;
6267 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6268 if (tso < 0) {
6269 dev_kfree_skb_any(skb);
6270 return NETDEV_TX_OK;
6271 }
6272
6273 if (tso)
6274 tx_flags |= IXGBE_TX_FLAGS_TSO;
6275 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6276 (skb->ip_summed == CHECKSUM_PARTIAL))
6277 tx_flags |= IXGBE_TX_FLAGS_CSUM;
6278 }
6279
6280 count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6281 if (count) {
6282 /* add the ATR filter if ATR is on */
6283 if (tx_ring->atr_sample_rate) {
6284 ++tx_ring->atr_count;
6285 if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6286 test_bit(__IXGBE_FDIR_INIT_DONE,
6287 &tx_ring->reinit_state)) {
6288 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6289 tx_flags);
6290 tx_ring->atr_count = 0;
6291 }
6292 }
6293 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6294 txq->tx_bytes += skb->len;
6295 txq->tx_packets++;
6296 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6297 hdr_len);
6298 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6299
6300 } else {
6301 dev_kfree_skb_any(skb);
6302 tx_ring->tx_buffer_info[first].time_stamp = 0;
6303 tx_ring->next_to_use = first;
6304 }
6305
6306 return NETDEV_TX_OK;
6307 }
6308
6309 /**
6310 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6311 * @netdev: network interface device structure
6312 * @p: pointer to an address structure
6313 *
6314 * Returns 0 on success, negative on failure
6315 **/
6316 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6317 {
6318 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6319 struct ixgbe_hw *hw = &adapter->hw;
6320 struct sockaddr *addr = p;
6321
6322 if (!is_valid_ether_addr(addr->sa_data))
6323 return -EADDRNOTAVAIL;
6324
6325 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6326 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6327
6328 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6329 IXGBE_RAH_AV);
6330
6331 return 0;
6332 }
6333
6334 static int
6335 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6336 {
6337 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6338 struct ixgbe_hw *hw = &adapter->hw;
6339 u16 value;
6340 int rc;
6341
6342 if (prtad != hw->phy.mdio.prtad)
6343 return -EINVAL;
6344 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6345 if (!rc)
6346 rc = value;
6347 return rc;
6348 }
6349
6350 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6351 u16 addr, u16 value)
6352 {
6353 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6354 struct ixgbe_hw *hw = &adapter->hw;
6355
6356 if (prtad != hw->phy.mdio.prtad)
6357 return -EINVAL;
6358 return hw->phy.ops.write_reg(hw, addr, devad, value);
6359 }
6360
6361 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6362 {
6363 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6364
6365 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6366 }
6367
6368 /**
6369 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6370 * netdev->dev_addrs
6371 * @netdev: network interface device structure
6372 *
6373 * Returns non-zero on failure
6374 **/
6375 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6376 {
6377 int err = 0;
6378 struct ixgbe_adapter *adapter = netdev_priv(dev);
6379 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6380
6381 if (is_valid_ether_addr(mac->san_addr)) {
6382 rtnl_lock();
6383 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6384 rtnl_unlock();
6385 }
6386 return err;
6387 }
6388
6389 /**
6390 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6391 * netdev->dev_addrs
6392 * @netdev: network interface device structure
6393 *
6394 * Returns non-zero on failure
6395 **/
6396 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6397 {
6398 int err = 0;
6399 struct ixgbe_adapter *adapter = netdev_priv(dev);
6400 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6401
6402 if (is_valid_ether_addr(mac->san_addr)) {
6403 rtnl_lock();
6404 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6405 rtnl_unlock();
6406 }
6407 return err;
6408 }
6409
6410 #ifdef CONFIG_NET_POLL_CONTROLLER
6411 /*
6412 * Polling 'interrupt' - used by things like netconsole to send skbs
6413 * without having to re-enable interrupts. It's not called while
6414 * the interrupt routine is executing.
6415 */
6416 static void ixgbe_netpoll(struct net_device *netdev)
6417 {
6418 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6419 int i;
6420
6421 /* if interface is down do nothing */
6422 if (test_bit(__IXGBE_DOWN, &adapter->state))
6423 return;
6424
6425 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6426 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6427 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6428 for (i = 0; i < num_q_vectors; i++) {
6429 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6430 ixgbe_msix_clean_many(0, q_vector);
6431 }
6432 } else {
6433 ixgbe_intr(adapter->pdev->irq, netdev);
6434 }
6435 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6436 }
6437 #endif
6438
6439 static const struct net_device_ops ixgbe_netdev_ops = {
6440 .ndo_open = ixgbe_open,
6441 .ndo_stop = ixgbe_close,
6442 .ndo_start_xmit = ixgbe_xmit_frame,
6443 .ndo_select_queue = ixgbe_select_queue,
6444 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6445 .ndo_set_multicast_list = ixgbe_set_rx_mode,
6446 .ndo_validate_addr = eth_validate_addr,
6447 .ndo_set_mac_address = ixgbe_set_mac,
6448 .ndo_change_mtu = ixgbe_change_mtu,
6449 .ndo_tx_timeout = ixgbe_tx_timeout,
6450 .ndo_vlan_rx_register = ixgbe_vlan_rx_register,
6451 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6452 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6453 .ndo_do_ioctl = ixgbe_ioctl,
6454 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6455 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6456 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6457 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6458 #ifdef CONFIG_NET_POLL_CONTROLLER
6459 .ndo_poll_controller = ixgbe_netpoll,
6460 #endif
6461 #ifdef IXGBE_FCOE
6462 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6463 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6464 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6465 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6466 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6467 #endif /* IXGBE_FCOE */
6468 };
6469
6470 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6471 const struct ixgbe_info *ii)
6472 {
6473 #ifdef CONFIG_PCI_IOV
6474 struct ixgbe_hw *hw = &adapter->hw;
6475 int err;
6476
6477 if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6478 return;
6479
6480 /* The 82599 supports up to 64 VFs per physical function
6481 * but this implementation limits allocation to 63 so that
6482 * basic networking resources are still available to the
6483 * physical function
6484 */
6485 adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6486 adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6487 err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6488 if (err) {
6489 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6490 goto err_novfs;
6491 }
6492 /* If call to enable VFs succeeded then allocate memory
6493 * for per VF control structures.
6494 */
6495 adapter->vfinfo =
6496 kcalloc(adapter->num_vfs,
6497 sizeof(struct vf_data_storage), GFP_KERNEL);
6498 if (adapter->vfinfo) {
6499 /* Now that we're sure SR-IOV is enabled
6500 * and memory allocated set up the mailbox parameters
6501 */
6502 ixgbe_init_mbx_params_pf(hw);
6503 memcpy(&hw->mbx.ops, ii->mbx_ops,
6504 sizeof(hw->mbx.ops));
6505
6506 /* Disable RSC when in SR-IOV mode */
6507 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6508 IXGBE_FLAG2_RSC_ENABLED);
6509 return;
6510 }
6511
6512 /* Oh oh */
6513 e_err(probe, "Unable to allocate memory for VF Data Storage - "
6514 "SRIOV disabled\n");
6515 pci_disable_sriov(adapter->pdev);
6516
6517 err_novfs:
6518 adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6519 adapter->num_vfs = 0;
6520 #endif /* CONFIG_PCI_IOV */
6521 }
6522
6523 /**
6524 * ixgbe_probe - Device Initialization Routine
6525 * @pdev: PCI device information struct
6526 * @ent: entry in ixgbe_pci_tbl
6527 *
6528 * Returns 0 on success, negative on failure
6529 *
6530 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6531 * The OS initialization, configuring of the adapter private structure,
6532 * and a hardware reset occur.
6533 **/
6534 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6535 const struct pci_device_id *ent)
6536 {
6537 struct net_device *netdev;
6538 struct ixgbe_adapter *adapter = NULL;
6539 struct ixgbe_hw *hw;
6540 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6541 static int cards_found;
6542 int i, err, pci_using_dac;
6543 unsigned int indices = num_possible_cpus();
6544 #ifdef IXGBE_FCOE
6545 u16 device_caps;
6546 #endif
6547 u32 part_num, eec;
6548
6549 /* Catch broken hardware that put the wrong VF device ID in
6550 * the PCIe SR-IOV capability.
6551 */
6552 if (pdev->is_virtfn) {
6553 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6554 pci_name(pdev), pdev->vendor, pdev->device);
6555 return -EINVAL;
6556 }
6557
6558 err = pci_enable_device_mem(pdev);
6559 if (err)
6560 return err;
6561
6562 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6563 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6564 pci_using_dac = 1;
6565 } else {
6566 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6567 if (err) {
6568 err = dma_set_coherent_mask(&pdev->dev,
6569 DMA_BIT_MASK(32));
6570 if (err) {
6571 dev_err(&pdev->dev,
6572 "No usable DMA configuration, aborting\n");
6573 goto err_dma;
6574 }
6575 }
6576 pci_using_dac = 0;
6577 }
6578
6579 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6580 IORESOURCE_MEM), ixgbe_driver_name);
6581 if (err) {
6582 dev_err(&pdev->dev,
6583 "pci_request_selected_regions failed 0x%x\n", err);
6584 goto err_pci_reg;
6585 }
6586
6587 pci_enable_pcie_error_reporting(pdev);
6588
6589 pci_set_master(pdev);
6590 pci_save_state(pdev);
6591
6592 if (ii->mac == ixgbe_mac_82598EB)
6593 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6594 else
6595 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6596
6597 indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6598 #ifdef IXGBE_FCOE
6599 indices += min_t(unsigned int, num_possible_cpus(),
6600 IXGBE_MAX_FCOE_INDICES);
6601 #endif
6602 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6603 if (!netdev) {
6604 err = -ENOMEM;
6605 goto err_alloc_etherdev;
6606 }
6607
6608 SET_NETDEV_DEV(netdev, &pdev->dev);
6609
6610 pci_set_drvdata(pdev, netdev);
6611 adapter = netdev_priv(netdev);
6612
6613 adapter->netdev = netdev;
6614 adapter->pdev = pdev;
6615 hw = &adapter->hw;
6616 hw->back = adapter;
6617 adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6618
6619 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6620 pci_resource_len(pdev, 0));
6621 if (!hw->hw_addr) {
6622 err = -EIO;
6623 goto err_ioremap;
6624 }
6625
6626 for (i = 1; i <= 5; i++) {
6627 if (pci_resource_len(pdev, i) == 0)
6628 continue;
6629 }
6630
6631 netdev->netdev_ops = &ixgbe_netdev_ops;
6632 ixgbe_set_ethtool_ops(netdev);
6633 netdev->watchdog_timeo = 5 * HZ;
6634 strcpy(netdev->name, pci_name(pdev));
6635
6636 adapter->bd_number = cards_found;
6637
6638 /* Setup hw api */
6639 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6640 hw->mac.type = ii->mac;
6641
6642 /* EEPROM */
6643 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6644 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6645 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6646 if (!(eec & (1 << 8)))
6647 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6648
6649 /* PHY */
6650 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6651 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6652 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6653 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6654 hw->phy.mdio.mmds = 0;
6655 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6656 hw->phy.mdio.dev = netdev;
6657 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6658 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6659
6660 /* set up this timer and work struct before calling get_invariants
6661 * which might start the timer
6662 */
6663 init_timer(&adapter->sfp_timer);
6664 adapter->sfp_timer.function = &ixgbe_sfp_timer;
6665 adapter->sfp_timer.data = (unsigned long) adapter;
6666
6667 INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6668
6669 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6670 INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6671
6672 /* a new SFP+ module arrival, called from GPI SDP2 context */
6673 INIT_WORK(&adapter->sfp_config_module_task,
6674 ixgbe_sfp_config_module_task);
6675
6676 ii->get_invariants(hw);
6677
6678 /* setup the private structure */
6679 err = ixgbe_sw_init(adapter);
6680 if (err)
6681 goto err_sw_init;
6682
6683 /* Make it possible the adapter to be woken up via WOL */
6684 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6685 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6686
6687 /*
6688 * If there is a fan on this device and it has failed log the
6689 * failure.
6690 */
6691 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6692 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6693 if (esdp & IXGBE_ESDP_SDP1)
6694 e_crit(probe, "Fan has stopped, replace the adapter\n");
6695 }
6696
6697 /* reset_hw fills in the perm_addr as well */
6698 hw->phy.reset_if_overtemp = true;
6699 err = hw->mac.ops.reset_hw(hw);
6700 hw->phy.reset_if_overtemp = false;
6701 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6702 hw->mac.type == ixgbe_mac_82598EB) {
6703 /*
6704 * Start a kernel thread to watch for a module to arrive.
6705 * Only do this for 82598, since 82599 will generate
6706 * interrupts on module arrival.
6707 */
6708 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6709 mod_timer(&adapter->sfp_timer,
6710 round_jiffies(jiffies + (2 * HZ)));
6711 err = 0;
6712 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6713 e_dev_err("failed to initialize because an unsupported SFP+ "
6714 "module type was detected.\n");
6715 e_dev_err("Reload the driver after installing a supported "
6716 "module.\n");
6717 goto err_sw_init;
6718 } else if (err) {
6719 e_dev_err("HW Init failed: %d\n", err);
6720 goto err_sw_init;
6721 }
6722
6723 ixgbe_probe_vf(adapter, ii);
6724
6725 netdev->features = NETIF_F_SG |
6726 NETIF_F_IP_CSUM |
6727 NETIF_F_HW_VLAN_TX |
6728 NETIF_F_HW_VLAN_RX |
6729 NETIF_F_HW_VLAN_FILTER;
6730
6731 netdev->features |= NETIF_F_IPV6_CSUM;
6732 netdev->features |= NETIF_F_TSO;
6733 netdev->features |= NETIF_F_TSO6;
6734 netdev->features |= NETIF_F_GRO;
6735
6736 if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6737 netdev->features |= NETIF_F_SCTP_CSUM;
6738
6739 netdev->vlan_features |= NETIF_F_TSO;
6740 netdev->vlan_features |= NETIF_F_TSO6;
6741 netdev->vlan_features |= NETIF_F_IP_CSUM;
6742 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6743 netdev->vlan_features |= NETIF_F_SG;
6744
6745 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6746 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6747 IXGBE_FLAG_DCB_ENABLED);
6748 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6749 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6750
6751 #ifdef CONFIG_IXGBE_DCB
6752 netdev->dcbnl_ops = &dcbnl_ops;
6753 #endif
6754
6755 #ifdef IXGBE_FCOE
6756 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6757 if (hw->mac.ops.get_device_caps) {
6758 hw->mac.ops.get_device_caps(hw, &device_caps);
6759 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6760 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6761 }
6762 }
6763 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6764 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6765 netdev->vlan_features |= NETIF_F_FSO;
6766 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6767 }
6768 #endif /* IXGBE_FCOE */
6769 if (pci_using_dac)
6770 netdev->features |= NETIF_F_HIGHDMA;
6771
6772 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6773 netdev->features |= NETIF_F_LRO;
6774
6775 /* make sure the EEPROM is good */
6776 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6777 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6778 err = -EIO;
6779 goto err_eeprom;
6780 }
6781
6782 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6783 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6784
6785 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6786 e_dev_err("invalid MAC address\n");
6787 err = -EIO;
6788 goto err_eeprom;
6789 }
6790
6791 /* power down the optics */
6792 if (hw->phy.multispeed_fiber)
6793 hw->mac.ops.disable_tx_laser(hw);
6794
6795 init_timer(&adapter->watchdog_timer);
6796 adapter->watchdog_timer.function = &ixgbe_watchdog;
6797 adapter->watchdog_timer.data = (unsigned long)adapter;
6798
6799 INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6800 INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6801
6802 err = ixgbe_init_interrupt_scheme(adapter);
6803 if (err)
6804 goto err_sw_init;
6805
6806 switch (pdev->device) {
6807 case IXGBE_DEV_ID_82599_KX4:
6808 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6809 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6810 break;
6811 default:
6812 adapter->wol = 0;
6813 break;
6814 }
6815 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6816
6817 /* pick up the PCI bus settings for reporting later */
6818 hw->mac.ops.get_bus_info(hw);
6819
6820 /* print bus type/speed/width info */
6821 e_dev_info("(PCI Express:%s:%s) %pM\n",
6822 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
6823 (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
6824 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
6825 (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
6826 (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
6827 "Unknown"),
6828 netdev->dev_addr);
6829 ixgbe_read_pba_num_generic(hw, &part_num);
6830 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6831 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6832 "PBA No: %06x-%03x\n",
6833 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6834 (part_num >> 8), (part_num & 0xff));
6835 else
6836 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6837 hw->mac.type, hw->phy.type,
6838 (part_num >> 8), (part_num & 0xff));
6839
6840 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6841 e_dev_warn("PCI-Express bandwidth available for this card is "
6842 "not sufficient for optimal performance.\n");
6843 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6844 "is required.\n");
6845 }
6846
6847 /* save off EEPROM version number */
6848 hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6849
6850 /* reset the hardware with the new settings */
6851 err = hw->mac.ops.start_hw(hw);
6852
6853 if (err == IXGBE_ERR_EEPROM_VERSION) {
6854 /* We are running on a pre-production device, log a warning */
6855 e_dev_warn("This device is a pre-production adapter/LOM. "
6856 "Please be aware there may be issues associated "
6857 "with your hardware. If you are experiencing "
6858 "problems please contact your Intel or hardware "
6859 "representative who provided you with this "
6860 "hardware.\n");
6861 }
6862 strcpy(netdev->name, "eth%d");
6863 err = register_netdev(netdev);
6864 if (err)
6865 goto err_register;
6866
6867 /* carrier off reporting is important to ethtool even BEFORE open */
6868 netif_carrier_off(netdev);
6869
6870 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6871 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6872 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6873
6874 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6875 INIT_WORK(&adapter->check_overtemp_task, ixgbe_check_overtemp_task);
6876 #ifdef CONFIG_IXGBE_DCA
6877 if (dca_add_requester(&pdev->dev) == 0) {
6878 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6879 ixgbe_setup_dca(adapter);
6880 }
6881 #endif
6882 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6883 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
6884 for (i = 0; i < adapter->num_vfs; i++)
6885 ixgbe_vf_configuration(pdev, (i | 0x10000000));
6886 }
6887
6888 /* add san mac addr to netdev */
6889 ixgbe_add_sanmac_netdev(netdev);
6890
6891 e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6892 cards_found++;
6893 return 0;
6894
6895 err_register:
6896 ixgbe_release_hw_control(adapter);
6897 ixgbe_clear_interrupt_scheme(adapter);
6898 err_sw_init:
6899 err_eeprom:
6900 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6901 ixgbe_disable_sriov(adapter);
6902 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6903 del_timer_sync(&adapter->sfp_timer);
6904 cancel_work_sync(&adapter->sfp_task);
6905 cancel_work_sync(&adapter->multispeed_fiber_task);
6906 cancel_work_sync(&adapter->sfp_config_module_task);
6907 iounmap(hw->hw_addr);
6908 err_ioremap:
6909 free_netdev(netdev);
6910 err_alloc_etherdev:
6911 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6912 IORESOURCE_MEM));
6913 err_pci_reg:
6914 err_dma:
6915 pci_disable_device(pdev);
6916 return err;
6917 }
6918
6919 /**
6920 * ixgbe_remove - Device Removal Routine
6921 * @pdev: PCI device information struct
6922 *
6923 * ixgbe_remove is called by the PCI subsystem to alert the driver
6924 * that it should release a PCI device. The could be caused by a
6925 * Hot-Plug event, or because the driver is going to be removed from
6926 * memory.
6927 **/
6928 static void __devexit ixgbe_remove(struct pci_dev *pdev)
6929 {
6930 struct net_device *netdev = pci_get_drvdata(pdev);
6931 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6932
6933 set_bit(__IXGBE_DOWN, &adapter->state);
6934 /* clear the module not found bit to make sure the worker won't
6935 * reschedule
6936 */
6937 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6938 del_timer_sync(&adapter->watchdog_timer);
6939
6940 del_timer_sync(&adapter->sfp_timer);
6941 cancel_work_sync(&adapter->watchdog_task);
6942 cancel_work_sync(&adapter->sfp_task);
6943 cancel_work_sync(&adapter->multispeed_fiber_task);
6944 cancel_work_sync(&adapter->sfp_config_module_task);
6945 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6946 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6947 cancel_work_sync(&adapter->fdir_reinit_task);
6948 flush_scheduled_work();
6949
6950 #ifdef CONFIG_IXGBE_DCA
6951 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
6952 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
6953 dca_remove_requester(&pdev->dev);
6954 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
6955 }
6956
6957 #endif
6958 #ifdef IXGBE_FCOE
6959 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
6960 ixgbe_cleanup_fcoe(adapter);
6961
6962 #endif /* IXGBE_FCOE */
6963
6964 /* remove the added san mac */
6965 ixgbe_del_sanmac_netdev(netdev);
6966
6967 if (netdev->reg_state == NETREG_REGISTERED)
6968 unregister_netdev(netdev);
6969
6970 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6971 ixgbe_disable_sriov(adapter);
6972
6973 ixgbe_clear_interrupt_scheme(adapter);
6974
6975 ixgbe_release_hw_control(adapter);
6976
6977 iounmap(adapter->hw.hw_addr);
6978 pci_release_selected_regions(pdev, pci_select_bars(pdev,
6979 IORESOURCE_MEM));
6980
6981 e_dev_info("complete\n");
6982
6983 free_netdev(netdev);
6984
6985 pci_disable_pcie_error_reporting(pdev);
6986
6987 pci_disable_device(pdev);
6988 }
6989
6990 /**
6991 * ixgbe_io_error_detected - called when PCI error is detected
6992 * @pdev: Pointer to PCI device
6993 * @state: The current pci connection state
6994 *
6995 * This function is called after a PCI bus error affecting
6996 * this device has been detected.
6997 */
6998 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
6999 pci_channel_state_t state)
7000 {
7001 struct net_device *netdev = pci_get_drvdata(pdev);
7002 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7003
7004 netif_device_detach(netdev);
7005
7006 if (state == pci_channel_io_perm_failure)
7007 return PCI_ERS_RESULT_DISCONNECT;
7008
7009 if (netif_running(netdev))
7010 ixgbe_down(adapter);
7011 pci_disable_device(pdev);
7012
7013 /* Request a slot reset. */
7014 return PCI_ERS_RESULT_NEED_RESET;
7015 }
7016
7017 /**
7018 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7019 * @pdev: Pointer to PCI device
7020 *
7021 * Restart the card from scratch, as if from a cold-boot.
7022 */
7023 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7024 {
7025 struct net_device *netdev = pci_get_drvdata(pdev);
7026 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7027 pci_ers_result_t result;
7028 int err;
7029
7030 if (pci_enable_device_mem(pdev)) {
7031 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7032 result = PCI_ERS_RESULT_DISCONNECT;
7033 } else {
7034 pci_set_master(pdev);
7035 pci_restore_state(pdev);
7036 pci_save_state(pdev);
7037
7038 pci_wake_from_d3(pdev, false);
7039
7040 ixgbe_reset(adapter);
7041 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7042 result = PCI_ERS_RESULT_RECOVERED;
7043 }
7044
7045 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7046 if (err) {
7047 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7048 "failed 0x%0x\n", err);
7049 /* non-fatal, continue */
7050 }
7051
7052 return result;
7053 }
7054
7055 /**
7056 * ixgbe_io_resume - called when traffic can start flowing again.
7057 * @pdev: Pointer to PCI device
7058 *
7059 * This callback is called when the error recovery driver tells us that
7060 * its OK to resume normal operation.
7061 */
7062 static void ixgbe_io_resume(struct pci_dev *pdev)
7063 {
7064 struct net_device *netdev = pci_get_drvdata(pdev);
7065 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7066
7067 if (netif_running(netdev)) {
7068 if (ixgbe_up(adapter)) {
7069 e_info(probe, "ixgbe_up failed after reset\n");
7070 return;
7071 }
7072 }
7073
7074 netif_device_attach(netdev);
7075 }
7076
7077 static struct pci_error_handlers ixgbe_err_handler = {
7078 .error_detected = ixgbe_io_error_detected,
7079 .slot_reset = ixgbe_io_slot_reset,
7080 .resume = ixgbe_io_resume,
7081 };
7082
7083 static struct pci_driver ixgbe_driver = {
7084 .name = ixgbe_driver_name,
7085 .id_table = ixgbe_pci_tbl,
7086 .probe = ixgbe_probe,
7087 .remove = __devexit_p(ixgbe_remove),
7088 #ifdef CONFIG_PM
7089 .suspend = ixgbe_suspend,
7090 .resume = ixgbe_resume,
7091 #endif
7092 .shutdown = ixgbe_shutdown,
7093 .err_handler = &ixgbe_err_handler
7094 };
7095
7096 /**
7097 * ixgbe_init_module - Driver Registration Routine
7098 *
7099 * ixgbe_init_module is the first routine called when the driver is
7100 * loaded. All it does is register with the PCI subsystem.
7101 **/
7102 static int __init ixgbe_init_module(void)
7103 {
7104 int ret;
7105 pr_info("%s - version %s\n", ixgbe_driver_string,
7106 ixgbe_driver_version);
7107 pr_info("%s\n", ixgbe_copyright);
7108
7109 #ifdef CONFIG_IXGBE_DCA
7110 dca_register_notify(&dca_notifier);
7111 #endif
7112
7113 ret = pci_register_driver(&ixgbe_driver);
7114 return ret;
7115 }
7116
7117 module_init(ixgbe_init_module);
7118
7119 /**
7120 * ixgbe_exit_module - Driver Exit Cleanup Routine
7121 *
7122 * ixgbe_exit_module is called just before the driver is removed
7123 * from memory.
7124 **/
7125 static void __exit ixgbe_exit_module(void)
7126 {
7127 #ifdef CONFIG_IXGBE_DCA
7128 dca_unregister_notify(&dca_notifier);
7129 #endif
7130 pci_unregister_driver(&ixgbe_driver);
7131 }
7132
7133 #ifdef CONFIG_IXGBE_DCA
7134 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7135 void *p)
7136 {
7137 int ret_val;
7138
7139 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7140 __ixgbe_notify_dca);
7141
7142 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7143 }
7144
7145 #endif /* CONFIG_IXGBE_DCA */
7146
7147 /**
7148 * ixgbe_get_hw_dev return device
7149 * used by hardware layer to print debugging information
7150 **/
7151 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7152 {
7153 struct ixgbe_adapter *adapter = hw->back;
7154 return adapter->netdev;
7155 }
7156
7157 module_exit(ixgbe_exit_module);
7158
7159 /* ixgbe_main.c */
Linux kernel - Deliverables
This page took 0.193664 seconds and 5 git commands to generate.